Nowadays, data is growing very speedy. Analysis of the data is necessity for the many organization. As per recent survey data generated in last 2 years is more than the data created in entire previous history of human. Data created in different form and in diversified manner. It can be structured, it can be semi-structured, or it can be unstructured. To analyze diversified by agricultural data we can use the tools of Big Data like Pig. Using Pig, we can analyze varieties of data. Pig is a platform for analysis of data. Biggest advantage of Pig is it can process any diversified data very quickly and allows us to use user defined functions. Use Case of Pig is ETL. It is used to extract the data from sources then after applying transformation we can load it into the data warehouse.
Here, in this study we analyzed verities of agricultural data using the big data tools Pig.
What is Pig?
Figure 1: What is Pig? |
Why Pig? & What Pig Supports?
Figure 2: Why Pig? & What Pig Supports? |
Analysis of Structured Agricultural Data Using Pig
To analyze structured data, first we must identify the source of data. Source of structured data can be any RDBMS like oracle, SQL Server, DB2, MySQL, Spreadsheets or OLTP Systems. Following are the source of structured data.
Figure 3: Sources of Structured Data |
Step-1 Load the structured data.
We took the data of state wise proportion circulation of Numeral of operative properties for all societal collections in 2005-06 and 2010-11 from government website [1].
Once retrieve the comma separated values file from government website, we copied the file on linux platform. Once we copied on linux then we moved the same file on HDFS platform. Following is command to move the file from linux root directory to HDFS directory named PARAG. CopyFromLocal command is used to move the file from linux directory to HDFS directory.
hadoop fs -copyFromLocal /root/state_data.csv /PARAG
Figure 3a |
Step-2 Display the loaded data
We can use dump statement to display the data in Grunt Shell.
Figure 4: State wise proportion circulation of Numeral of operative properties |
Step-3 Filter Specific Data
For analysis of any data we can use filter or aggregate functions. Here, we are filtering the specific data from state Gujarat.
Figure 5: State wise data of 2005 which census marginal is more than 50 |
Finding all state data which census_small of 2005 is more than 30
Figure 6: State wise data of 2010 which census marginal is more than 80 |
Finding all state data which census_small of 2010 is more than 30
Figure 7: State wise data of 2010 which census small is more than 30 |
Analysis of Unstructured Agricultural Data Using Pig
Conclusion
We did analysis of agricultural data of state wise proportion circulation of Numeral of operative properties for all societal collections in 2005-06 and 2010-11 using Pig. We analyzed structured agricultural data using Pig. As we know that day by day requirement of analysis of the data is increasing rapidly. To demonstrate the use of analysis using big data tools Pig we used the government agricultural data and did the analysis of data.
Analysis of the data is necessity for the many organization. Data created in different form and in diversified manner. It can be structured, it can be semi-structured, or it can be unstructured. To analyze diversified by agricultural data we can use the tools of Big Data like Pig. Using Pig, we can analyze varieties of data. Pig is a platform for analysis of data. Biggest advantage of Pig is it can process any diversified data very quickly and allows us to use user defined functions. Use Case of Pig is ETL. It is used to extract the data from sources then after applying transformation we can load it into the data warehouse.
Acknowledgment
We wish to thank Open Government Data Platform (OGD) for providing data for analysis & sincere thanks to our mentor.
References
The increased interest to theprogress of the computer engineering, computer and telecommunication networks can be observed lately. This caused the creation of large information resources, which are to be stored, processed and transferred. Therefore the tasks of protecting computer-based information become of great importance. These tasks are generally solved by means of deriving information by performing a cryptographic transformation. Recently the protection of information in computer systems is supported on the basis of knowledge about the biometrical features of human voice, signature, eye’s iris and other biometrical parameters.
A lot of methods and systems based on biometrical principles for reliable protection of fast changing information streams are developed lately. They include development of the effective methods of compression and recognition of speech signals.
The most of the known algorithms of determination of biometrical parameters features are based on the methods which are not enough accurate to describe the specific changes of signal (pulses, peaks etc). Therefore the received data contain only overall information about the signal. There is as a rule a great shortcoming of information to conduct the effective recognition. Through their variety the abovementioned algorithms adapt themselves badly to the analysis of speech signals.
There is a great necessity of designing new methods for reliable recognition on the basis of wavelet-packet analysis and speech signals compression using Karhunen-Loev transformation for development of the biometrical systems for real-time protection of information streams [1].
Result and Analysis
The specific compression methods deserve the attention. In them the orthogonal conversion of a signal segment is conducted with the next exception of a collection of the least by absolute value factors of expansion. The additional transfer of information about arrangement of factors is necessary. In the elementary case the transmission of this information is conducted with a bit sequence and for each factors of expansion just one bit needs to be allocated. Therefore the total number of bits of the additional information is equal to the size of the transformation window and factor of compression К is calculated with equation [2]:
Where B is a quantity of bits allocated for each factor; N is a quantity of selections in a segment of processing; U is a quantity of the transferred factors of expansion.
As a result two information flows are formed: the first one corresponds to the information stream about factors of expansion and the second one transfers the information about arrangement of these factors.
The results of the mathematical simulation of the speech signal compression and decompression system with the help of the application in the MATLAB environment (figure 1) have permitted to present graphics of dependence of signal/distortion (s/dis) ratio from factor of compression for segments of conversion in 64, 128, 256 samples.
Comparing the presented results it is possible to make a conclusion that at double increase of a size of conversion segment, for example from 64 up to 128 samples, it is possible to achieve only very small increase of quality of compression (in the best case 4 Db), and in some cases even the decrease of the quality.
The form of the graphs for all explored conversions is similar. The very positive property is decrease of a droop steepness of s/dis when increasing the compression factor value. Therefore even the small improvement of orthogonal functions base will lead to increase of all dependence and as a result – to appreciable increase of possible compression factor [3].
Figure 1: Dependence of the s/dis relation by the compression factor for the transformation segments: a) 64 samples; b) 128 samples; c) 256 samples |
Let’s compare the valid compression factor for orthogonal conversions having fixed the ratio of s/dis at the level of 30 dB [4]. It is often impossible to expose precisely 30 Db. Then with the help of linear interpolation between two adjacent values we conduct the elimination of conversions factors of compression factor determination. The results of calculations for different selections are presented in the table.
Table 1: Possible factors of compression for the explored orthogonal transformations at the window of transformation (64, 128, 256) of sample. |
At subjective comparison of qualities of speech signal transmission (by ear), the distortions under the usage of inclined and Adamar transformations look like broadband noise. That can be explained by their specific structure (the Adamar transformation- two-level digital functions; the inclined transformation – triangular-like form functions). When using more harmonic transformation – the sine, cosine and Karhunen-Love ones, the signal becomes unnaturally “metal” and under considerable increase of compression factor the unnaturalness increases and reaches the complete loss of clearness of perception of some sound fragments of a phrase. At an increase of the size of the transformation window distortions become more extended and capture a few sounds, but their level does not decrease and the clearness of the speech does not improve.
It is studied that the human perception of these distortions depends on the last segment of the system, namely from the signal-sound converter, because the distortions under the high-quality converter along with the good sound the distortions are clearly heard. Simple membrane converters are used mainly in telephony. Therefore there is a poor quality of speech playback and the subjective level of distortions essentially diminishes on this background.
In the research paper the method of characteristics determination of the speech signals recognition and the algorithm of nonlinear time normalization are offered and substantiated. Basing on the high performance of the wavelet-packet analysis, a conclusion is made about possibility of its effective application to speech signals recognition and compression.
In the wavelet-packet algorithm the rapid wavelet-transformation (RWT) in the form of frequency sequential splitting operation is applied both for low frequency and high-frequency (detailed) factors. In this case the wavelets of each next level are emerged from the wavelet of the previous level by splitting into two new wavelets:
where h_{n},g_{n},are the appropriate weight factors; t is the time.
New wavelets are also localized but on two-times broader interval. Accordingly the complete set of wavelet expansion functions iscalled the wavelet -packet.
Wavelet-packet transformation is adaptive [5,6]. This fact allows fitting more precisely to the features of signals by the choice of an appropriate tree of the optimal expansion form which provides a minimum quantity of wavelet-factors at the given accuracy of the signal reconstruction. Thus the informational redundancy and unnecessary details of signals are switching-off [7].
The estimation of information level of a collection of wavelet-factors is carried out by the entropy under which understands the next value:
where x,x_{n} are signal expansion in the n-node and previous one.
Any effort of factors averaging multiplies the entropy. During the analysis of tree the entropy of knots and its spited parts are calculated. If during splitting the node the entropy does not decrease there is no necessity of the subsequent expansion of this node (such nodes are called the last or terminal ones).
The searching of optimal terminal nodes for creation the recognition class can be divided into the following steps. The first step consists in definition of a list of all terminal nodes that are contained in the optimized wavelet-trees of each image of the properly recognized class. The second step is the comparison of terminal nodes of identical numbers for different images of the given class. Speed of change of amount of transitions through zero is used in a role of comparison criterion. It is necessary to mark a case when numbers of terminal nodes for different images of one class do not fit. In such case a comparison of terminal node with nodes that are not a terminal one and are stored in other images of the same class but the numbers of them are equal to the numbers of terminal nodes.
The third step is definition of node number from the given list (can be terminal one) that presents a given image in the class. The maximum fitting of value of the function of transitions through zero speed change with the same functions obtained for other images is the criterion of choice.
On the next step over the obtained function the packing of a volume range is carried out. In the role of operation the nonlinear one of taking the logarithm can be used.
To receive the recognition factors the operation of spectral analysis is used.
Where m is the amount of expansion factors, S(n) is the amplitude spectrum of the signal, N is a selection.
To conduct the temporal normalization a strain function is searched. The implementation of this function minimizes disagreement between the standard and new realizations of words. The two functions are searched more precisely:
value of segmenting function from the proper contours.
Segmenting function should characterize an aggregate modification of parameters of a speech signal that are used by it and it depends on two frames: current one and previous one. The energy distribution of a signal on frequency groups is used in a role of parameters of a speech signal.
The procedure of determination of distortion functions ω_{x} and ω_{y} implemented by the method of dynamic programming and enables carrying out the interior nonlinear time smoothing of implementations of words.
Knowing the distortion functions ω_{x}, ω_{y }we can for any area of standard realization of word to find the proper area of a new realization. We will apply it for separation of the new implementation of a word on sound dyads. Sound dyad is a transient from phoneme to phoneme, which maps reorganization of a means of an articulation. Unlike realization of phoneme, realization of a sound dyad is much less inclined to influence of a context and map correlation of adjacent phonemes of a speech stream. The centers of pseudo-steady areas of phonemes are boundaries of dyad. Thus, dyad consists of the second half of the first phoneme and the first half of the second phoneme.
The standard realization of the word is divided by sound dyads manually: numbers of α0,…αL frames are stressed. These frames are the centers of pseudo-steady areas of phonemes. Then the points nl, ∫ = 0,…,L such, that ω_{x}(nl) = αl are selected. Now, with the help of the function ω_{y }it is possible to spot numbers of b0,…bLframes, that are the centers of pseudo-steady areas of phonemes in the new realization of word: bl = ω_{y}(nl), l = 0,….L.
The resulted procedure allows passing from the comparison of realization of words to comparison of realization of sound dyads. For providing of identical quality of transmission, both loud and weak signals, the adaptive quantization of decomposed factors are used, which takes place due to the change of an amplification factor of link that precedes to the compressor, therefore together with decomposed factors there is the necessity to passan amplification factor k(n). The amount of bits on the decomposed factor is defined under the equation:
whereW(n)is average level of decomposed factors,
qmax isthe maximal quantity of bits on a factor,
ceil()the operation of rounding to greater whole (at the negative value gives a zero),
- consideration of sign bit.
The greatest possible value of the unit of decomposed factor is defined under the formula:
The amount of bits necessary for the transmission of complete aggregate of decomposed factors cannot decrease more than twice. In order to promote the factor of compression is used adaptive uniform quantization, in which the adapting will be carried out for all decomposed factors (segment of factors). We can really obtain the good results of compression using less than eight bits on factor.
The factor of compression at usage of this method is defined under the formula:
where K_{B}- quantity of bits selected for the transmission of amplification factor,
В – quantity of bits on a factor on the entrance of compressor,
K_{nul} – quantity of zero factors for which sign bits are not needed. For comparison of orthogonal conversions it is possible to use the complex quality index, but evidently it does not give the picture of such parameters as a factor of compression or correlation of s/dis, which is more better for comparison of compression with the usage of one of the orthogonal conversion.
On the figure 2 and figure 3 the value of factors of compression is compared for different transformations for the fixed relation of s/dis at the level of 30 dB.
Figure 2: Possible factors of compression for explored orthogonal transformations at the window of transformation (64 samples). |
The program testing methods of recognition is carried out by definition of probability of an classification error with the help of Mahalanobisdistance (Gonzales). The connection between Mahalanobis distance and identification error is presented with the next equation (Gonzales)[8]:
where p (e) is the probability of error r_{ij} is the Mahalanobis distance.
Figure 3: Possible factors of compression for explored orthogonal transformations at the window of transformation (128 samples). |
In the process of testing the 6 classes of images were used. One of them was the correct on. So, Mahalanobisdistance was calculated between the correct class and the five wrong classes. Estimation of error for the worst class makes 30%, and for the best one approximately 5%. Estimation of the average error of detection for all classes makes 20 %.
Figure 4: Dependence of error probability on Mahalanobis distance |
The selection in our case is smaller in the statistical value. In the general case the range of small selection as a result of many researches of number sequences makes from 10-15 to 200. There is no sense to hope for a regular of statistical performances of average value and variance of probability of a correct (wrong) recognition at such circumstances [9]. If to use deciding rule, which is founded on a principle of the nearest neighbor, in this case reliability of a correct (wrong) recognition is equal 1 (0). It is precisely known that the probability of a correct recognition is value less 1 for selections of arbitrary size. Nevertheless classical statistical approaches have no sufficient sensitivity that in condition of small selections reliably to avoid a single (zero) event in relation to probability of correct (wrong) recognition. The arbitrary classical statistical estimation in conditions of small selections is characterized by the large values of dispersions, which extremely strongly aggravate reliability of the most statistical estimation.
The methods of the differential estimation of reliability of correct (wrong) recognition are applied to avoid the problems of estimation with the help of the statistical approaches. Also it is proved that the differential reliability averaged behind the database of correct (wrong) recognition equals probabilities of correct (wrong) recognition of algorithm as a whole. Thus dispersion of the constructed estimation of reliability of a correct (wrong) recognition in some times can be smaller than in case of the classical statistical estimations. On completion it is necessary to mark that the estimation of the average reliability of wrong recognition at the level of 20 % responds to such class of biometric systems.
Conclusion
The new method of signals recognition features determination is developed with the use of wavelet-packet analysis is designed and the analysis of efficiency of methods of speech signals compression is conducted. The scientific and practical results are as following:
References
Data mining is that the method to extract or mine data from immense volume of information. Broadly data processing will be outlined because the task of extracting implicit, antecedently unknown potential helpful data from knowledge in giant databases. Data mining tasks are classified as descriptive which discover interesting patterns or relationships describing the data and predictive task which predicts or classifies the behavior of the model supported obtainable information. It’s a content field with a general goal of predicting outcomes and uncovering relationships. Some of the data mining techniques are Classification, Clustering and Rule Mining.
Clustering is that the most typically used information discovery technique. It helps un-covering the unknown category labels. It helps un-covering the unknown class labels. Clustering has gained importance in many applications in the recent past. Most of the cluster algorithms area unit ascendable to large dataset. Weather is random entity. Forecasting is the technology to predict the atmosphere at given location and a given time taking into consideration various factors such as humidity, temperature, wind and outlook.. It’s done by gathering the information regarding this state of the atmosphere at a given location thus applies scientific understanding to predict but the temperature will modification over the course of some time. In our paper we are going to predict whether the play can happen based on current weather values such as temperature, humidity, windy, outlook [11]. We make the prediction based on various classification algorithms such as Decision Tree (J48), REP Tree and Random Tree. We conjointly compare every of those algorithms in terms of their accuracy mistreatment completely different measures.
Classification algorithms
Decision Tree Induction
DTI is a tree learning algorithms. It consists of flow diagram like structure wherever the inner node denotes a take a look at on the attribute, the branches will denote the outcome of the test performed on the attribute and the leaf nodes will denote class labels.
The internal nodes are represented as rectangles and the leaf nodes are represented with oval shapes.
To determine the cacophonic attribute it makes use of various attribute choice measures like data gain, gain quantitative relation and Gini Index.
Example:J48,C 4.5,CART
REP Tree
It is a decision tree learner algorithm. It constructs the decision tree exploitation data gain or variance then prunes it exploitation reduced error pruning exploitation back fitting strategy.REP Tree Iteratively generates multiple trees using regression logic. It sorts the values for numeric attribute onlyonce. It deals with missing values by rending the corresponding instances into items.
Random Trees
This algorithm can deal with both regression and classification problems. it’s a group of tree predictors that’s referred to as forest. It takes the input as feature vector and compares it with each tree within the forest and offers the result category label that has highest votes.
Classifier Output Measures
The classifier classifies the tuples in the dataset. It is quite natural that the classifier may have error rate and may fail to correctly classify the tuples.Hence we measure the classifier accuracy which is given by the percentage of instances that square measure properly classified by classifier.
Confusion Matrix
It gives information regarding the classifier output in terms of the number of tuples that are correctly classified and the numbers of tuples that are miss classified. For a good accuracy classifier the elements must be in along the diagonal while the other entries must be close to zero.
Mean Absolute Error
It is a measure for accuracy. It is the mean of the absolute errors that is the mean of the distinction between the expected value and also the actual value.
Root Mean Square Error
If we take the square root of the mean square error then we obtain the root mean square error. We do it to adjust large error rates.
Results and Comparisons
The tool we used for the result analysis is WEKA which consists of large number of open source machine learning algorithms. It takes the input in the form of ARFF (Attribute Relation File Format),CSV(comma separated values).The data set we used is weather which is input to weka in ARFF format.
The weather data set contains following attributes.
Table 1 |
The dataset has 14 instances
Figure 1: Statistics of J48 classifier on weather dataset |
Result of REP Tree Classifier
Figure 2: Statistics of REP Tree classifier on weather dataset |
Result of Random Tree
Figure 3: Statistics of Random Tree classifier on weather dataset |
Overall comparison of J48, REP Tree and Random Tree (using training dataset)
Table 2 |
Conclusion
Thispaper intends to study the classifier accuracy of various classification algorithms using WEKA tool on weather dataset.The experimental results of the various classification algorithms is listed.First the experiment was done on the weather dataset using j48 algorithm which classifies all the instances correctly.The accuracy of the j48 classifier is 100%.Then the dataset was run on Random Tree classifier which classifies all instances correctly and has 100% accuracy.Then classification was done using REP Tree classifier and we found the accuracy was decreased to 64.28% because it was not able to classify all the instances correctly and we found that 5 instances were misclassified by REP Tree classifier because of which its accuracy is decreased.
References
The utilization report provided by Federal Communications Commission (FCC)^{1} spectrum task force indicates that utilization of the allocated spectrum varies between 15% to 85%. Mitola^{2,3} is considered to be the pioneer of the idea of cognitive radio (CR). CR^{4}is a smart device having the capability to sense the entire spectrum in order to find the idle channels and utilizing these idle channels opportunistically for communication as and when required.
The important issues of CR are as under:
Spectrum Sensing
This function finds the idle spectrum to be used for its communication and the arrival of PU on the spectrum channels presently occupied by CR.
Spectrum management
Large chunks of idle spectrum are identified through spectrum sensing that needs to be categorized so that the best spectrum channelis selected for CR’s communication.
Spectrum mobility
Primary user has the priority to access the spectrum. Once PU appears on the channels occupied by CR, it needs to change the operating frequency instantly to minimize the interference.
Spectrum sharing
Once transmitting channels are selected, handshaking process is initiated between communicating users first,before actual communication starts.
Providing QoS guarantees to different flows as per their requirement in CR is a very difficult job. QoS^{5} is the ability of the system to provide minimum requisite service needed for satisfactory communication between users and has direct impact over CR performance. It is also defined by the degree of satisfaction of the CR users and their skill to treat different flows differently^{6,7}. The definition of QoS changes with types of traffic and network. The QoS is a function of multiple contradictory parameters and those are not certain and exact. Application of fuzzy logic gives more effective solution when the available information is vague and imprecise. This paper proposes to use rule based fuzzy logic for assessing the influence of three critical parameters on QoS in a cognitive radio network. At the onset, fuzzy logic found applications in control system^{8,9} and then in telecommunication networks^{10-12}. Fuzzy logic found more applications in cellular networks for admission control^{13}and for routing^{14}. It was extended to CR network for transmitting power control^{15 }and foroptimization^{16}. It found more applications for controlling the spectrum access^{17,18}and spectrum handoff^{19-21 }in cognitive radio. The authors of^{22,23}have analysed impact of some QoS parameters.
Materials Aand Methods
An algorithm, using rule-based fuzzy inference system, is proposed in this paper to analyse the impact of three critical parameters namely signal-strength, bandwidth and user-mobility on the QoS in cognitive radio. We have used Hong and Rappaport traffic model^{24,25}for user-mobility. The assumptions of the model are:
The hand off requests ^{24,25}arrive into the system as
R_{n} is the probability that the communication at least need one more hand off without blocking the service request.
P_{hh} is the probability that the on-going communication need one more hand off.
B_{o} is the probability of blocking of new service requests.
P_{f} is the probability that handoff is not successful.
λ_{o} is the arrival rate of service requests in a cell.
Three inputsto fuzzy inference system are given as:
a) input 1: signal-strength (in dBm)
b) input 2: bandwidth (in KHz)
c) input 3: user-mobility (in km/hour)
and the output of the fuzzy inference system is the quality of the communication channel in CR.Each inputis characterized by three fuzzy sub-sets, such as low, medium and high,as shown in (2) defined by universe of discourse in the range[0,1]
S(signal-strength) = S(bandwidth) = S(user-mobility) = {low, medium, high}(2)
and QoS is characterized by five fuzzy sub-sets, such as poor, average, good, very-good and excellent,as shown in(3),definedbyuniverse of discoursein the range[0,1]
S(QoS)={poor, average, good, very-good,
excellent}(3)
The analysis of the proposed algorithm is being done using trapezoidal membership functions for inputs and the QoS,depicted in Figs. 1(a) and 1(b), and for normalized values in the range[0,1]. Table.a depicts the rules of the fuzzy inference system. The rules are designed to have the suitable values of the three contradictory inputs so as to satisfy the QoS requirement of the communicating users. For instance, rule 2demonstrates that if the signal-strength is low, unused bandwidth is low and user-mobility is medium, then the quality of the communicating channel is poor. The rule is translated into actual channel condition in which transmission loss of the communicating channel is very large, availability of unused bandwidth is limited and the user is moving with speeds that falls under middle category. As a result, the quality of the channel is not sufficient to satisfy QoS requirements of the communicating users. Rule 14demonstrates that if the signal-strength is medium, bandwidth is medium and user-mobility is medium, then the quality of the communicating channel is average. The rule is translated into actual channel condition in which transmission loss of the communication channel is in middle range, availability of unused bandwidth is sufficient and the user is moving with speeds that fall under middle category. As a result, the quality of the channel is just sufficient to satisfy QoS requirements of the communicating users.Also rule 26demonstrates that if the signal-strength is high, bandwidth is high and user-mobility is medium, then the quality of the communicating channel is excellent. The rule is translated into actual channel condition in which transmission loss of the communication channel is negligible, availability of unused bandwidth is very large and the user is moving with speeds that fall under middle category. As a result, the quality of the channel is more than sufficient so that the QoS requirement of the communicating users is easily satisfied.Fig.2 shows the decision of fuzzy inference system for inputs such as signal strength = 0.5 (middle value), bandwidth = 0.5 (middle value), user-mobility = 0 (stationary) and for these values the QoS comes out to be 0.5 (excellent).
Figure 1: Membership function plots of the inputs a)signal-strength, bandwidth and user-mobility and consequent b) QoS |
Table 1: Rulesin fuzzy inference system |
Figure 2: Decision of fuzzy inference system for QoS |
Results and Discussions
The proposed algorithm is analysed for static data transfers in Matlab’sfuzzy toolbox. Figs. 3(a), 3(b) and 3(c)demonstrate the effect of signal-strength, bandwidth and user-mobility on the quality of the communication between users while keeping the other two parameters constant at middle values (=0.5).The results indicate that with increase in the signal-strength there is increase in the quality of the communication of the users, while keeping bandwidth and user-mobility constant at middle values, as shown in Fig. 3(a). The further results indicate that with the increase in available unused bandwidth of a communication channel, there is also an increase in the quality of the communication of the users, while keeping signal-strength and user-mobility constant at middle values, as shown in Fig. 3(b). Fig. 3(c) indicates that with the increase inuser-mobility i.e. when the speed of the user increases, there is decrease in the quality of the communication of the users,while keeping bandwidth and signal-strength constant at middle values.
Figure 3: Analytical results show effect of a) signal-strength onQoS b) bandwidth on QoS c) user-mobility on QoS while keeping other two parameters constant at middle values (= 0.5) |
Figs. 4(a), 4(b) and 4(c)demonstrate the effect of bandwidth and signal-strength,user-mobilityand signal-strength, and user-mobilityand bandwidth on the quality of the communication between users while keeping the other parameter constant at middle value (=0.5).Decision surface results indicate that the quality of the communicating channels increases with increase in signal-strength and the availability of unused bandwidth, when the user is moving with constant medium speed, as shown in Fig. 4(a). The results indicate that the QoSin the communicating channel is the maximal, when the unused bandwidth and signal-strength,areat their maximal values. Decision surface results indicate that the quality of the communicating channel increases with increase in signal-strength and decrease in speed of the user while the available unused bandwidth is kept constant at middle value, as shown in Fig. 4(b). The results indicate that the QoSin the communicating channel is the maximal, when signal-strength is at its maximal value and user remains stationary or moves with very low speed.Decision surface results indicate that the quality of the communicating channel increases with increase in the availability of unused bandwidth and decrease in speed of the user, while the signal-strength is kept constant at middle value, as shown in Fig. 4(c). The results indicate that the QoS in the communicating channel is the maximal,when the availability of unused bandwidth is maximal and user remains stationary or moves with very low speed.
Figure 4: Decision surface showing the effect of a) signal-strength and bandwidth on QoS b) user-mobility and signal- strength on QoS and c) user-mobility and bandwidth on QoS while keeping other parameter constant at middle value (= 0.5) |
Conclusion
The paper proposed an algorithm using fuzzy inference system that assesses the impact of signal-strength, bandwidth and user-mobility on QoS in cognitive radio. The analysis of the proposed algorithm is being performed in Matlab’s fuzzy toolbox. The results indicate that with increase in the signal-strength and unused bandwidth, and decrease in user-mobility, there is increase in QoS of the communication between CR’s.
References
Missing data imputation techniques can be used to improve the data quality. Missing data imputation techniques replace missing values of a dataset so that data analysis methods can be applied to complete dataset [1].
In this research work, student dataset is taken contains marks of four different subjects in engineering college. Mean, Mode, Median Imputation were used to deal with challenges of incomplete data. By using MSE and RMSE on dataset using with proposed Method and imputation methods like Mean, Mode, and Median Imputation on the dataset and found out to be values of Mean Squared Error and Root Mean Squared Error for the dataset. Accuracy also found out to be using Proposed Method with Imputation Technique. Experimental observation it was found that, MSE and RMSE gradually decreases when size of the databases is gradually increases by using proposed Method. Also MSE and RMSE gradually increase when size of the databases is gradually increases by using simple imputation technique. Accuracy is also increases with increases size of the databases.
The organization of the paper is Section1: Introduction Section 2: Literature Reviews, Section 3: Dataset Used, Section 4: Methodology, Section 5:Experimental Result and Analysis, Section 6: Conclusions
Literature Reviews
lit wise Deletion
This method deletes those instances with missing value for data analysis in the dataset. It is the most common method, it has two drawbacks: a) A substantially decreases the size of dataset available for the data analysis. b) Data are not always missing completely at random.
Mean/Mode Imputation (MMI)
By replacing a missing values with the mean or mode of all attribute which consists missing value. To reduce the influence of exceptional data, median can also be used. This is one of the most commonly used methods.
K-Nearest Neighbor Imputation (KNN)
This method uses k-nearest neighbor algorithms to estimate and replace missing data. The main advantages of this method are a) it can estimate both qualitative attributes and quantitative attributes; b) It is not necessary to build a predictive model for each attribute with missing data [2].
Median Substitution
Median Substitution is calculated by grouping up of data and finding average for the data. Median can be calculated by using the formula
Median =L+h/f (n/2-c) (1)
where L is the lower class boundary of median class h is the size of median class i.e. difference between upper and lower class boundaries of median class f is the frequency of median class, c is previous cumulative frequency of the median class, n/2 is total no. of observations divided by 2
Standard Deviation
The standard deviation measures the spread of the data about the mean value. It is useful in comparing sets of data which may have the same mean but a different range. The Standard Deviation is given by the formula
Where{X1,X2,…,X_{n}}are the observed values of the sample items and is the mean value of these observations, while the denominator N stands for the size of the sample [7].
Dataset Used
In this work dataset having characteristics is given below.
Number of Instances: 5000,10,000,15,000,20,000
Number of Attributes: 05
(Record No., M1, ECE, EM, EE)
Dataset contains marks of four different subjects of engineering college. In dataset randomly distributed the missing values in each attribute to become the incomplete dataset. Record. No. in the Dataset is used are imaginary and generated for the data analysis purpose in data mining process. In dataset M1, ECE, EM, EE are the subject in engineering college and class test marks for each subject is out of twenty marks for each subject repectively.The structure of Dataset as shown in the Table No.1
Table 1: Dataset |
Methodology
4.1to found out accuracy by using proposed method with imputation technique like mean, mode and median for five thousand dataset, ten thousand dataset, fifteen thousand dataset and twenty thousand dataset.
4.2 To found out MSE (Mean Squared Error) for proposed method with mean imputation technique and simple mean imputation technique by using following equation
4.3 To found out MSE (Mean Squared Error) for proposed method with mode imputation technique and simple mode imputation technique by using equation no.1.
4.4. To found out MSE (Mean Squared Error) for proposed method with median imputation technique and simple median imputation technique by using equation no.1.
4.5 To found out RMSE (Root Mean Squared Error) for proposed method with mean imputation technique and simple mean imputation technique by using following equation
4.6 To found out RMSE (Root Mean Squared Error) for proposed method with mode imputation technique and simple mode imputation technique by using equation no.2
4.7 To found out RMSE (Root Mean Squared Error) for proposed method with median imputation technique and simple median imputation technique by using equation no.2
Experimental Result and Analysis
5.1 For Experimental Result Student dataset is taken which contains marks of four different subjects of engineering college. Mean, Mode, Median Imputation were used to deal with challenges of incomplete data.
5.2 Using proposed method with imputation techniques like Mean, Mode, and Median Imputation on the student dataset and found out to be accuracy for five thousand, ten thousand, fifteen thousand and twenty thousand dataset respectively.Accuracy increase with increase in size of the dataset.
5.3 Similarly found out to be MSE by using Proposed Method with Imputation Technique like Mean, Mode, and Median Imputation.MSE decreases with increase in size of the dataset.
5.4 Found out to be MSE by using simple Imputation Technique like Mean, Mode, and Median Imputation. MSE increase with increase in size of the dataset.
5.5 Similarly found out to be RMSE by using Proposed Method with Imputation Technique like Mean, Mode, and Median Imputation. RMSE decreases with increase in size of the dataset.
5.6 Also found out to be RMSE by using simple Imputation Technique like Mean, Mode, and Median Imputation. RMSE increase with increase in size of the dataset.
5.7 Accuracy found out to be by proposed method with Imputation Technique like Mean, Mode, and Median Imputation. Also MSE by using Proposed Method with Imputation Technique and MSE by using simple Imputation Technique, RMSE by using Proposed Method with Imputation Technique, RMSE by using simple Imputation Technique. The result is shown in the Table no.2 and Table no.
Table 2: Comparison of Accuracy and MSE |
Figure 1: Graphical Representation of Accuracy and MSE |
Table 3: Comparison of Accuracy and RMSE |
Figure 2: Graphical Representations of Accuracy and RMSE |
Conclusions
In this research paper by using proposed method with imputation technique like Mean, Mode and Median Imputation on the student dataset and found out to be accuracy. Also found out to be MSE and RMSE on dataset using with proposed Method and imputation methods like Mean, Mode, and Median Imputation on the dataset. Experimental observation it is found that, MSE and RMSE gradually decreases when size of the databases is gradually increases by using proposed Method. Also MSE and RMSE gradually increase when size of the databases is gradually increases by using simple imputation technique. Accuracy increase with increase in size of the dataset.
Acknowledgment
I wishes to thank my guide and Data Analytics Laboratory of Department of Computer Science and Information Technology, Dr. Babasaheb Ambedkar Marathwada University, Aurangabad.
References
The idea of Fibonacci cordial labeling was given by A. H. Rokad and G. V. Ghodasara[1]. The graphs which i considered here are Simple, undirected, connected and finite. Here V(G) and E(G) denotes the set of vertices and set of edges of a graph G respectively. For different graph theoretic symbols and nomenclature i refer Gross and Yellen[3]. A dynamic survey of labeling of graphs is released and modified every year by Gallian[4].
Definition 1
Let G = (V(G), E(G)) be a graph with V = X_{1}∪X_{2}∪X_{3}∪. . . Xi∪Y
where each X i is a set of vertices having at least two vertices of the same degree
and Y=V\∪Xi.The degree splitting graph of G designated by DS (G) is acquired from G by adding vertices z_{1},z_{2}, z_{3},. . . , z_{y }and joining to each vertex of xi for i £[1, t].
Definition 2
The double fan DFn comprises of two fan graph that have a common path. In other words DFn = Pn+ K_{2 }
Definition3
The duplication of an edgee= xy of graph Gproduces a new graph G’by adding an edgee’= x’y ’such that N(x’)=N(x)∪{y’}−{y}and N(y’)=N(y)∪{x’}−{x}.
Definition4
The graph obtained by connecting a vertex of first copy of a graph G with a vertex of second copy of a graph G is called joint sum of two copies of G.
Definition5
A globe is a graph obtained from two isolated vertex are joined by n paths of length two.It is denoted byGl(n).
Definition 6
Ring sum G_{1}⊕ G_{2}of two graphs G_{1}= (V_{1}, E_{1}) and G_{2}= (V_{2}, E_{2}) is the graph G_{1}⊕ G_{2}= (V_{1}∪ V_{2}, (E_{1}∪ E_{2}) − (E_{1}∩ E_{2})).
Results
Theorem 1
DS(Pn) is Fibonacci cordial.
Proof 1
Consider Pn with V (Pn) = {v_{i}: i [1, n]}. Here V (Pn) = X_{1}∪ X_{2}, where X_{1}= {x_{i}: 2 [2, n-1]} and X_{2}= {x_{1}, x_{n}}. To get DS(Pn)from G we add w_{1 }and w_{2 }corresponding to X_{1}andX_{2}.Then
|V(DS(Pn))|=n+2andE(DS(Pn))={X_{1}w_{2}, X_{2}w_{2}}∪{w_{1}x_{i}:i [2,n–1]}. So,|E(DS(Pn))|=−1 + 2n.
I determine labeling function g: V(G) → {F_{0}, F_{1}, F_{2}, . . . , F_{n+2}} as below:
g(w_{1})= F_{1},
g (w_{2})= F_{n+1},
g(x_{1})= F_{0},
g(x_{i}) = F_{i}, 2 ≤ i ≤ n.
Therefore,|e_{g}(1)−e_{g}(0)|≤1.
Therefore, DS(Pn) is a Fibonacci cordial graph.
Example 1
Fibonacci cordial labeling of DS(P_{7}) can be seen in Figure 1.
Figure 1 |
Theorem 2
DS(DFn) is a Fibonacci cordial graph.
Proof2
Let G=Dfn be the double fan. Letx_{1}, x _{2},…, x _{n}be the path vertices of Dfn and x and y be two apex vertices. To get DS(Dfn) from G,we add w_{1}, w_{2 }corresponding to X_{1}, X_{2},where
X_{1}={x_{i}: i [1, n] }and X_{2}= {x, y}.Then|V(DS(Dfn))| = 4+ n&E(DS(Dfn)) ={xw_{2}, y w_{2}}∪{x_{i}w_{1}: i [1, n] }. So,|E(DS(Dfn))|= 1+ 4n.
I determine labeling function g:V(G)→{F_{0},F_{1},F_{2},…,F_{n+4}},as below.
For all 1 ≤ i ≤n.
g (w_{1}) = F_{3},
g(w_{2}) = F_{2}.
g(x) = F_{0},
g (y) = F_{1},
g(x_{i}) = F_{i+3}.
Therefore |e_{g}(1)−e_{g}(0)|≤1.
Therefore, DS(DFn) is Fibonacci cordial.
Example 2
Fibonacci cordial labeling of DS(DF_{5}) can be seen in Figure 2
Figure 2 |
Theorem 3
The graph obtained by duplication of an edge in K_{1, n}is a Fibonacci cordial graph.
Proof3
Let x_{0 }be the apex vertex and x _{1}, x _{2},…, x _{n}be the consecutive pendant verticesofK_{1, n}.Let G be the graph obtained by duplication of the edgee= x _{0} x _{n }by an e wedge e’= x_{0}’x_{n}’. There for einG,deg(x_{0}) = n,deg(x_{0}’) = n,deg(v_{n}) = 1, deg(x_{n}’) = 1anddeg(x_{i}) = 2,∀ i∈{1,2,…n}. Then|V(K_{1, n})| =n+3 and E(K_{1, n}) =2n.
I determine labeling function g: V(G) → {F_{0}, F_{1}, F_{2}, . . . , F_{n+3}}, as below.
g(x_{0})= F_{1},
g (x_{1}) = F_{2},
g(x_{n−1})=F_{3},
g(x_{0}’)=F_{0},
g(x_{n}’)=F_{4},
g(x_{i})=F_{i}_{+3}, i [2,n],in−1.
Therefore |e_{g}(1)−e_{g}(0)|≤1.
Therefore,the graph obtained by duplication of an edge in K_{1, n}is a Fibonacci cordial graph.
Example 3
A Fibonacci cordial labeling of the graph obtained by duplication of an edge e in K_{1, 8}can be seen in the Figure 3.
Figure 3 |
Theorem4
The graph obtained by joint sum of two copies of Globe Gl(n)is Fibonacci cordial.
Proof4
Let G be the joint sum of two copies of Gl (n).Let{x, x’, x_{1}, x _{2},…, x _{n}} and
{y, y ’, y_{1}, y _{2},…, y _{n}} be the vertices of first and second copy of Gl(n)respectively.
I determine labeling function g: V(G) → {F_{0}, F_{1}, F_{2}, . . . , F_{2n+4}}, as below.
g(x) = F_{0},
g(x’)= F_{1},
g(x_{i})=F_{i+3},i [1, n].
g(y)=F_{2},
g(y’)= F_{3},
g(y_{i}) = F_{n+i+3}, i [1, n].
From the above labeling pattern i have eg (0)=n+1ande_{g}(1)=n.
Therefore |e_{g}(1)−e_{g}(0)|≤1.
Thus, the graph obtained by joint sum of two copies of Globe G l(n)is Fibonacci cordial.
Example 4
Fibonacci cordial labeling of the joint sum of two copies of Globe Gl(7) can be seen in Figure 4
Figure 4 |
Theorem 5
The graph DF_{n}⊕ K_{1, n}is a Fibonacci cordial graph for every n ∈ N .
Proof 5
Assume V(DF_{n}⊕K_{1, n})= X_{1}∪ X _{2}, where X_{1}={x,w, x _{1}, x _{2},…, x _{n}}be the vertex set of DF_{n }and X_{2}={y=w, y _{1}, y _{2},…, y _{n}} i sthevertexsetofK_{1, n}. Here v is the apex vertex & y _{1}, y _{2},…, y _{n }are pendant vertices of K_{1, n}.
Also |V (DFn ⊕ K_{1,n})| = 2n + 2, |E(DFn ⊕ K_{1,n})| = 4n − 1.
I determine labeling function g:V(DF_{n}⊕K_{1, n})→{F_{0},F_{1},F_{2},…,F_{2n+2}},as below.
Forall1≤i≤n.
g (x)=F_{0},
g (w)=F_{1},
g(x_{i}) = F_{i+1},
g (y_{i}) = F_{n+i+1},
From the above labeling pattern i have e_{g}(0) = 2n and e_{g}(1) = 2n−1.
Therefore||e _{g}(1)−e_{g}(0)|≤1.
Thus, the graph DF_{n}⊕ K_{1, n}is a Fibonacci cordial graph for every n ∈ N .
Example 5
Fibonacci cordial labeling of DF_{5}⊕ K_{1, 5}can be seen in Figure 5.
Figure 5 |
Theorem 6
The graph G ⊕ K_{1, n}is a Fibonacci cordial graph for all n ≥4, n∈N,where G is the cycle C_{n }with one chord forming a triangle with two edges ofC_{n}.
Proof 6
Let G be the cycle C_{n}with one chord. Let V (G ⊕ K_{1, n}) = X_{1}∪ X _{2}, where X _{1}isthevertexsetofG& X _{2}isthevertexsetofK_{1, n}. Let x _{1}, x _{2},…, x _{n}be the successive vertices of C_{n }and e= x _{2 }x_{n }be the chord of C_{n}.The vertices x_{1}, x _{2}, x _{n }forma triangle with the chorde.Here v is the a pexvertex & y _{1}, y _{2},…, y _{n}are pendant verticesofK_{1, n}.
I determine labeling function g: V (G ⊕K_{1, n}) → {F_{0}, F_{1},F_{2},. . . , F_{2n}}, as below.
CaseI:n≡0(mod3).
For all 1 ≤ i ≤ n.
g (x_{i}) =F_{i}.
g(y_{i}) = F_{n+i}.
CaseII:n≡1(mod3).
g (x_{i}) = F_{i}, 1 ≤ i ≤ n.
g (y_{1}) = F_{0},
g(y_{i}) = F_{n+i−1}, 2 ≤ i ≤ n.
From the above labeling pattern i have e_{g}(0)=n and e_{g}(1)=n+1.
Therefore |e_{g}(1)−e_{g}(0)|≤1.
Thus, the graph G ⊕ K_{1, n}is a Fibonacci cordial graph.
Example 6
A Fibonacci cordial labeling of ring sum of C_{7}with one chord and
K_{1, 7}can be seen in Figure 6.
Figure 6 |
Theorem 7
The graph G ⊕ K_{1, n}is a Fibonacci cordial graph for all n ≥5, n∈N,where G is the cycle with twin chords forming two triangles and another cycleC_{n−2}with the edges of C_{n}.
Proof7
Let G be the cycle C_{n }with twin chords,where chords form two triangles and one cycle C_{n−2}.LetV(G⊕K_{1, n})=X_{1}∪ X _{2}. X_{1}={x_{1}, x _{2},…, x _{n}} is the vertex set of C_{n}, e_{1}= x_{n }x_{2}and e_{2}= x_{n }x_{3} are the chords of C_{n}. X_{2}={y= x_{1}, y_{1}, y_{2},…, y_{n}}is the vertex set of K_{1, n},where y_{1}, y_{2},…, y_{n }are pendant vertices and y= x_{1}istheapexvertexofK_{1, n}.Also|V(G⊕K_{1, n})|=2n, |E(G ⊕ K_{1, n})| = 2n + 2.
Take y= x _{1}.Also|V(G⊕K_{1, n})|=2n, |E(G ⊕ K1,n)| = 2n + 1.
I determine labeling function g: V(G ⊕K_{1, n}) → {F_{0}, F_{1},F_{2},. . . , F_{2n}}, as below.
g (x_{1})= F_{1},
g(x_{2})= F_{2},
g (x_{3})= F_{3},
g (x_{n}) = F_{4},
g (x_{i}) = F_{i+1}, 4 ≤ i ≤ n − 1.
g (y_{i}) = F_{n+i}, 1 ≤ i ≤ n.
From the above labeling pattern i have e_{g }(0)=e_{g}(1)=n+1.
Therefore |e_{g}(1)−e_{g}(0)|≤1.
Thus, The graph G ⊕ K_{1, n}is a Fibonacci cordial graph.
Example7
A Fibonacci cordial labeling of ring sum of C_{9 }with twin chords andK_{1, 9}can be seen in Figure7.
Figure 7 |
Conclusion
In this paper i investigate seven new graph which admits Fibonacci cordial labeling.
References
Technology nowadays is overwhelming that it has results change faster from a stand-alone system to a web based technology which is capable of supporting almost all of the computerized transactions using an open source mobile applet and Content Management Systems. Most of the organizations have embraced technology and have developed exceptional online programs that provide easy access and massive communication. Philippine Merchant Marine Academy (PMMA) is one of the premier maritime institutions whose advocacy is to produce competent maritime professionals and the only maritime school in the province of Zambales. The Philippine Merchant Marine Academy (PMMA) is currently divided into two colleges namely; the College of Bachelor of Science in Marine Engineering, and the College of Bachelor of Science in Marine Transportation. From the total of 290 2^{nd} class cadets, 150 are B.S. Marine Engineering cadets and 140 are B.S. Marine who are currently having their on –board ship internship training in preparation for their fields of specialization as deck and engine officer.
A web-based Applet Monitoring System via Cloud-Based Application using Android and Joomla Technology can provide efficient and accurate monitoring of performance reports of the 2^{nd} class cadets. This intends to perform the following operations; analyze and transfer the gathered reports and eliminate circumstances of causing the delay of releasing reports; minimize the time consumed in searching for the name of the cadets in the filing cabinet and in the spreadsheet program. Web-Based and Android Applet has to be applied out since web-based and mobile technology takes place in most of the industry nowadays because of its accessibility. It is expected that the computerization or the application of the information system provides the technical basis to develop sound managerial decisions.[1],[2],[3],[4],[5],[6],[7],[8],[9]
Using the power of internet technology the researcher proposes to design and develop a cloud-based performance monitoring system which may ease up the flow of the old system and will help reduce the use of papers in encoding and validating the performance of the cadets.
In the present scenario, the retrieval of the report uses the manual-based training record book which causes inconsistency and late submission of reports when it comes to monitoring the performance of the cadets while on board ship training. This is because other shipping companies send their reports after the training period of the cadets. In the proposed monitoring system the summary of report on the performance of the cadets will be directly encoded in the Web Based system by the training officers of the shipping agency. At the same time, the Cadets can also receive the summary of their performance through online notification using the Android Applet. [1],[2],[3],[4].
Materials and Methods
Sampling Technique
The researcher has use a purposive sampling technique. The researcher will focus on particular characteristic of a population that are to answer the research questions.
The Respondents
The respondents of the study are the 2^{nd} class cadets and officers of the Department of Shipboard Training (DST) of the Philippine Merchant Marine Academy, San Narciso, Zambales and the officers of the various shipping companies. The respondents were officially enrolled during 2014-2015.
Systems Development and Life Cycle Is a calculated model utilized as a part of undertaking administration that portrays the stages engaged with a data framework improvement venture, from an underlying possibility ponder through upkeep of the finished application
Figure 1: shows the diagram of the Systems Development and Life Cycle (SDLC |
Figure 2: shows the context diagram of the Cloud Based Application. |
Gantt Chart
Figure 3: shows the schedules of activities of the development of the Cloud Based Application. |
Development
The researcher had developed a Cloudbased application system using PHP and MySql in order to automate the submission of sea projects done by the cadets and will be monitored by the deck and engine training officers.
A Gantt chart was also used in the development process in order to track down the progress of each phases.
Research Instrument
Questionnaire has used as a research instrument. The researcher has prepare structured questionnaire for the respondents to be used as a testing phase of the proposed system. The questionnaire consists of two parts. The first part is the respondents’ as to age, sex, course and occupation. The second part is the respondents’ perception in the existing system and proposed system with regards to the specified parameters such as accuracy, security, maintainability, operability, and speed of processing.
Validation of Research Instrument
Draft of questionnaire has been made by the researcher. Errors and modifications will be corrected by the thesis adviser to further improve the questionnaire. Final copy of the questionnaire will be distributed to the respondents.
Data Gathering Method
After answering the questionnaire, it will be immediately retrieved by the researcher from the respondents. The researcher tally the data that were collected from the respondents to were able to compute, analyze and interpret the result that were used in the presentation and analysis of data.
Results and Discussions
From the findings, the following conclusions which are binding on the respondents are arrived at:
Recommendations
In view of the findings and conclusions, the researcher would like to recommend the following:
Table 1: Frequency and Percentage Distribution of the Respondents in Terms of Course |
As seen from the table, out of 131 respondents from the existing group, 63 or 48.9 % of the respondents are enrolled BSMT , while 62 or 48.1% are BSMARE cadets and Officials are 4 or 3%. On the other hand, for the proposed group, out of the 139 respondents, 65 or 46.8% are BSMT, 66 or 47.5% are enrolled in BSMARE and Officials are 8 or 5.8 %, . As implied, the number of respondents as group per course is in parallel to the ratio of the students currently enrolled at PMMA.
Table 2: Test of Difference between the Existing System and Proposed Systems as group in terms of System Quality Metrics |
The researcher would like to express his sincere heartfelt and deepest gratitude to those who have rendered their time, knowledge, guidance and support to make this work a reality.
Dr. Nemia M. Galang, his very supportive adviser, for her encouragement, precious time and wisdom.
Dr. Menchie A. Dela Cruz, his efficient critic, for her patience, dedication and extended assistance to this study.
Dr. Esmen M. Cabal & Ms. Melojean C. Marave for their encouragement and sincere comments.
Dr. Domingo C. Edaño, the Director of Graduate Studies for giving encouragement to the researcher in pursuing this study.
MR. RAMON G. LACBAIN II, Former Zambales Vice-Governor for approving and giving him a financial assistance in pursuing his study.
VADM RICHARD U RITUAL PMMA, Superintendent of the Philippine Merchant Marine Academy, for his consideration, understanding and support.
Engr, Patrick Entendez, Director of Shipboard Training for his accommodation and support in the administration of his questionnaire and providing the data needed by the researcher, but moreover, for encouraging him to pursue his graduate studies.
His Friends, Marijoy S. Pagal, C/E Charlie M. Pandongan, Engr. Arnold Pasamba, PSITE Region III Officers, Dr. Dave Marcial, Dr. Grace Tyler of Systems Plus College Foundation, Engr. Alessandro Ong, Ms. Jessa Delos Santos Ladringan, Rev. Father Willie Monsalud, Rev. Dexter Magno, Knights of Columbus Council 10103 and my AMA Olongapo Family.
To his cousins Vivian Delute and Rebecca Lugatiman in giving their support and guidance in this study.
To his loving and supportive family, for their prayers, support and encouraging words.
To all his friends in Philippine Merchant Marine Academy and PMMA’s 1^{st} Class Midshipmen Batch 2014, who had assisted in distributing the questionnaires.
And above all, to God Almighty, for showering the writer all the love, guidance and blessings.
References
The spectrum is a valuable electromagnetic resource that is controlled by government in order to manage complex issues. Presently, spectrum is allocated to service providers by adopting the policy of fixed allocation in which transmission power is regulated and different frequency bands are assigned for different services and applications. There has been tremendous growth of wireless users and applications over the last decade. The total number of users worldwide was 3.2 billion in 2009 and it was projected to increase by 100 folds by 2013^{1}. These new applications require more spectrum allocation, but it is difficult to find the new spectrum as most of the spectrum bands are allocated by a policy of fixed allocation. This policy has created a situation where there appears an artificial scarcity of the spectrum. The spectrum utilization report provided by Federal Communications Commission (FCC)^{2} spectrum task force indicates that utilization of the allocated spectrum varies between 15% to 85% and is a function of space and time. Other spectrum occupancy measurements conducted have revealed that the average utilization in New York City^{3} was at 5.2% while in Chicago^{4} at 17.4%. Several other countries spectrum occupancy measurement studies, such as Spain^{5}, Singapore^{6}, Germany^{7,8}, New Zealand^{9} and United Kingdom^{10 }also confirmed that spectrum is heavily under-utilized . The authors of ^{11} have proposed the framework for future spectrum occupancy measurements covering the frequency range from 700 to 2700 MHz in India. Thus, the persistent increase in demand of spectrum cannot be fulfilled until a new scheme is not found to control the limited spectrum. This new scheme is the dynamic spectrum access (DSA)^{12-14} in which secondary users (SUs) can use the idle licensed channels opportunistically known as spectrum hole or white space with only constraint of minimum interference to primary users (PUs) or licensed users. If spectrum is utilized on time or frequency basis, the spectrum opportunities appear in the form of holes as shown in Fig.1. In time domain, it is the period of time during which the PU in not transmitting and in frequency domain, it is the frequency band in which SU can use the frequency band for its transmission allotted to PU with no or minimum interference. The technology that will make the DSA a reality is the Cognitive Radio (CR). CR is a smart device having the capability to sense the entire spectrum in order to find the idle channels and utilizing these idle channels opportunistically for communication as and when required. The opportunistic usage of these idle channels can increase the utilization of precious spectrum. CR can enable the SUs to utilize licensed channels of PUs either on negotiated or opportunistic basis. It can provide a basis for efficient wireless communication between users wherein nodes have the capability to alter any of its transmission or reception parameters in order to adapt to continuously changing environment.
Figure 1 |
Capabilities and Features
The smart device like CR can have the capabilities to model its location and time varying environment so that the suitable frequency channels, protocols and interfaces are selected in order to have the best communication. Cognitive cycle proposed by Mitola^{15} is shown in Fig. 2.
The proposed cognition cycle is described as under:
Observe
CR has the capability to sense the environment so that it has complete knowledge about the environment.
Orient
The importance of the collected information is assessed.
Plan
On the basis of collected information, CR determines all its available options for resource optimization.
Decide
Amongst the available options, the best course of action is decided.
Figure 2 |
Act
The best course of action is implemented by CR for resource optimizations. The changes are then reflected in the interference profile by CR.
Learn
During this process, the CR uses all its previous observations and decisions to improve upon its future decisions using learning techniques.
Spectrum As A Resource
Spectrum is a group of electromagnetic radiations used for wireless communications. Chunks of these different ranges of frequencies are allocated for different applications for dedicated, successful and secure service provisioning. Different spectrum bands are allocated for different services and applications in order to guarantee interference free operation of the users. But, fixed allocation of the spectrum bands has created poor utilization scenario of allocated bands that has resulted in wastage of valuable natural resource. With the increase in wireless applications and its users, the service providers have been at loggerhead with each other for acquiring more spectrum. The regulatory agencies have faced a tough time for spectrum distribution especially during past one decade. Every nation has its right for unbounded use of the spectrum. At the international level, International Telecommunication Union (ITU) has been assigned the responsibility of allocation of spectrum for various countries in world radio communication conference. Considering the difficulties in spectrum allocation, authorities like Federal Communications Commission (FCC), Telecom Regulatory Authority of India (TRAI) emphasizes the need for change in spectrum regulatory policy from fixed to dynamic spectrum allocation.
Spectrum Allocation in India: Present Scenario
In order to facilitate the hardware compatibility, standardization; the specific spectrum bands are allocated for a particular use. Allocation of different spectrum bands for different services and applications in India^{16 }is shown in Table 1.
Table 1 |
Government of India under ministry of communications created the wireless planning and co-ordination wing (WPC) in 1952 that acts as the National Radio Regulatory Authority. WPC is being assigned the responsibility of management of spectrum in India. Standing Advisory Committee on Frequency Allocation (SACFA) is one of the section of WPC. Main function of SACFA is to make recommendations regarding allocation issues and sort out other issues referred to it by wireless users. Wireless communication refers to the transmission of electromagnetic waves with range from 3 Hz to 300 GHz. The electromagnetic waves transmitted at different frequency ranges have different characteristics, such as interference, path loss, wireless link errors, link layer delay etc., necessitating the use of unique applications for specific frequency bands.
Issues of Cognitive Radio
The main purpose of the CR is to realize dynamic access to the idle spectrum in order to have communication. However, implementation of a cognitive radio is a challenging task. A typical cognitive radio scenario consists of SUs that co-exist with the PUs. PU has a priority to use the spectrum as they have legacy rights for the spectrum access. SU opportunistically access the spectrum when they find that PU is not using the spectrum or underlay access to the spectrum, when both PU and SU co-exist, with strict constraint over non-interference among the users. The main issues of CR are as under^{14,17,18}:
Spectrum Sensing
Spectrum sensing is the important function of the CR. The CR (or SU) has to sense the radio environment in order to detect the idle channels (or spectrum hole) and use these idle channels for its communication. The sensing operation should also be able to detect the PU’s arrival on the frequency band presently occupied by SUs instantly and accurately^{19}. This function can be further divided as under:
Primary transmitter detection
The CRs should have the capabilities to detect even weak signals from any primary transmitter through its local observations. This function can be further divided as under:
Matched filter detection
In this scheme, the CR has the prior knowledge of the some characteristics of the PU signal and is able to detect the presence or absence of PU with high accuracy^{20}. However, the disadvantage is the requirement of knowledge of some features of primary signal such as operating frequency, modulation scheme, packet format etc^{19}.
Energy detector
In this scheme, the CR does not have prior knowledge of any characteristics of the PU’s signal. The measured energy of the received signal is compared with the preselected threshold in order to determine the presence or absence of the primary signal^{21}. The disadvantage of this scheme is the non-discrimination between noise energy and primary user signal^{22}.
Cyclostationary feature detection
In this scheme, the modulated signals are coupled with other waves such as sine wave carriers, pulse trains, repeated spreading, hopping sequences or cyclic prefixes to create built-in periodicity. These cyclostationary features are detected at the receiver in order to determine the presence or absence of the primary signal. The advantage of this approach is the discrimination between the noise energy and the primary user signal^{23,24}. This is a complex technique and takes long time for detection of the PU signal.
Cooperative spectrum sensing
In this scheme, the local sensing result from multiple CR users is utilized to draw conclusion on the presence or absence of primary user signal. This approach mitigates the harmful effects of fading, shadowing, noise, interference etc. and results in reduction of miss detection and false alarm probabilities^{25,26}.
Interference based detection
In this scheme, the transmission of CR is allowed only if the interference introduced does not exceed a certain threshold known as interference temperature limit of the licensed receiver^{27}.
Spectrum Management
Spectrum management helps in finding the best idle channels or hole for the transmission of SU among large number of available spectrum holes. The SU has to interact with different protocol layers of the network to maintain the quality of service (QoS) of an application. Based on the availability of the spectrum, the channel is allocated to the SU. This decision also depends on the internal and external policies as well as other regulatory issues of the spectrum. This function can be further divided as under:
Spectrum analysis
The spectrum analysis performs the function of the characterization of the idle spectrum or hole found through the spectrum sensing. The characterization of these holes is necessary due to availability of very wide range of spectrum. The characteristics, such as interference, path loss, wireless link errors, link layer delay etc., varies with change in operating frequencies^{14}.
Spectrum decision
As per the requirement of the CR user, the appropriate spectrum band is selected for the transmission that satisfies the QoS of the SU. An opportunistic frequency channel skipping protocol^{28} is proposed that searches better quality channel based on SNR.
Spectrum Mobility
Spectrum mobility is the process of switching the spectrum band during data transmission due to arrival of PU on that band. The SU has to switch to another frequency band which is not used by the PU at that time in order to continue the transmission^{29}. The switching to this idle band should be seamless so that there is minimum QoS degradation of the application running over the SU. As soon as the PU needs the frequency band, the SU has to terminate its transmission and free the frequency band for the PU functioning^{30}.
Spectrum Sharing
Spectrum sharing is an important functionality of CR as it coordinates the traffic between secondary and primary users. It is a challenging task as it requires high degree of cooperation, understanding and coordination between primary and secondary users. Resource allocation and spectrum access are its two main functions. This function can be further divided as under:
On the basis of Architecture
Centralized spectrum sharing
In this type of sharing, a central entity controls resource allocation and spectrum access functions. A comparative study of centralized and distributed approaches indicates that distributed approaches generally follow the centralized approaches having disadvantage of message exchange^{31}.
Distributed spectrum sharing
In this type of sharing, CR itself manages the functions of resource allocation and spectrum access based on local (or global) policies^{32-34}.
Spectrum allocation behaviour
Cooperative spectrum sharing
This type of sharing considers the effect of CR’s communication on others. In this case, information is shared among the CRs or among the CRs and the central entity in order to take a decision^{32,33,35} .
Non- cooperative spectrum sharing
This type of sharing considers only the communication of a particular CR and thus, adopts a selfish approach. In this case, no information exchange takes place among CRs^{34,36}.
Spectrum access technique
Overlay spectrum sharing
In this type of sharing, the CR accesses the spectrum that is not being used by primary user at that time^{32,34,35,37}.
Underlay spectrum sharing
In this type of sharing, both CR and PU transmit in the same frequency band but the transmission of CR is considered noise by the primary user and thus, the interference introduced should be within the tolerable limits. Underlay systems use ultra-wide band^{38} or spread spectrum technique^{31 }for communication.
Conclusion
The present regulatory policies and devices working on fixed spectrum allocation are not suitable for rising service demands and are responsible for under-utilization of spectrum bands. Thus, there is a need to devise newer way of spectrum allocation strategies and new devices that will work efficiently based on dynamic spectrum allocation.The mentioned issues need to be addressed before a practical cognitive radio network is realised.
References
Mainly, Islam has five main pillars to believe in. Hajj to Mecca is the fifth pillar. Hajj is an annual Islamic physical religious journey to Mecca which is obligatory once in life time for the Muslim believers who are physically and financially capable perform it. Furthermore, Umrah is another physical religious journey to Mecca which can be performed at any time of the year. In Kingdom of Saudi Arabia (KSA) there are two holy cities for Muslim believers. Mecca is first holy city Muslims believers where both Umrah and Hajj are performed. Madinah is the second holy city in KSA for Muslim believers. People who are performing both Hajj and Umrah are called pilgrims. pilgrim: pilgrim: pilgrim: According to the Saudi General Authority for Statistics (https://www.stats.gov.sa), in average there are 10 million people entered Mecca from inside and outside the KSA as pilgrims in order to perform Umrah or Hajj yearly. In average, there are five million pilgrims performing Umran in the holy month of Ramdhan, and three million pilgrims are performing Hajj in the holy month of Dhu al-HijjahThe other two million pilgrims are performing Umrah in the other ten months of the year. According to new vision of the Saudi Government, the number of pilgrims who are looking to perform Umrah or Hajj will be increased to be 20 million pilgrims yearly[1][2].
According to the above facts, A huge crowd will occur in both cities Mecca and Madinah because of the pilgrims who are performing Umrah or Hajj. Because of this crowd, there are several questions can be asked such as: How to identify the dead, lost pilgrims? Also, How to gather the missing items and distributes it? The answer for such questions is to develop a recognition system to identify and detect people individually. The system will use a registered database of faces for all local and foreign pilgrims. A solution for such crowd is to design a robust recognition system in order to identify the pilgrims. A robust recognition system needs a huge database to be built. Such a huge database gives the robustness for any recognition system if well trained and tested. Working on an existing database for pilgrims is quiet difficult since the pilgrims are changing every year[1][2][3].
In order to perform Hajj or Umra, a foreign pilgrim who lives outside KSA needs a Hajj/Umrah visa to enter KSA. This visa takes place in all KSA embassies and consulates. On the other hand, local Umrah pilgrims can go to Mecca at any time of the year. But local Hajj pilgrims need a permission to perform Hajj by applying for it. The Mistry of Hajj in KSA checks about the eligibility for a pilgrim for the eligibility of a expatriate for Hajj. If the pilgrim is eligible for Hajj, he/she will be given a permit otherwise no permit will be issued. For local pilgrims, they are eligible to perform Hajj every five years [1][2][3].
Performing Hajj needs a registration via Umra/Hajj travel agent who deals with the KSA embassies and consulates all over the world for foreign pilgrims. Also, there are local Umra/Hajj travel agents who deal with the Ministry of Hajj [1][2].
Identifying the missed people framework system
In this paper, a new framework system for identifying the missed pilgrims in Hajj and Umrah is proposed. The system consists of two main phases. Firstly, Build the database is accomplished. Secondly, a recognition system is designed to identify the missed pilgrims. The recognition system is designed based or iris recognition. Preparing Database
Based on the fact, the pilgrims are not same pilgrims every year. Snicethe pilgrims are changing every year, a yearly huge database for all pilgrims is needed every year. Build such a database is done as follows:
Figure 1: Hajj ID Sample |
It is noted that the Hajj ID contains all the personal information for the pilgrim. This information is the pilgrim name, local address, both Madinah and Mecca residency, and the agent name.
Millions of pilgrims go to Mecca and Madinah in KSA to conduct Hajj or Umrah. Unfortunately, some of the pilgrims will miss their group member or they will die Iidentifying and finding the missed or dead pilgrims is a difficult [1][3].
There many application for identifying the missed pilgrims such as the e bracelet application. Special electronic (E) bracelets were issued for all pilgrims. The E bracelets contain all the details about the pilgrim wearer. Figure 2 shows a sample of the E bracelets. A drawback for such application is the there is a high probability to lose the E bracelets from the pilgrim wearer. The loss of the e bracelets may occur any time while washing the hands, or while taking a shower even though all the E bracelets are water resistant [1]. To overcome such a drawback, a new framework for identifying the missed pilgrims is proposed in this paper.
Figure 2: The E Hajj bracelet sample |
Proposed system
In this paper, a new framework for identifying the missed pilgrims in Hajj and Umrah is proposed. Basically, this system uses the conventional phases for a recognition system. The phases are summarizes as, image acquisition for preparing the database, the pre-processing for noise removal and normalization, the feature extraction for converting the 2D image into 1D vector, and the classification Phase for identification process [4]. The proposed system is shown in figure 3.
Figure 3: The Proposed System |
Preparing Database
The database for all pilgrims has to be gathered. Here, collecting iris images for all the pilgrims is the main goal. This is done in previous research by acquiring both iris and Hajj ID images in order to construct a huge database for pilgrims. The database for all pilgrims is stored in a huge storage at the Ministry of Hajj. Each year this database is updated due to different pilgrims are come to perform Hajj. For example the database for 2105 differs from the database for 2016.All the pre-processing techniques were conducted to the database in order to make it ready for classification and recognition [5].
The first phase in constructing and building a robust recognition system database is to find suitable source of data. In this paper, collecting the iris image for all pilgrims is the main challenge. In our previous research, a database for identifying the missed pilgrims in both Hajj and Umrah was proposed [6]. The iris image for all the foreign pilgrims who are coming to KSA via all the borders will be captured and saved in the Ministry of Hajj computer server. Also, capturing the iris image for the local pilgrims will be taken and saved as well.
In this paper, the system is proposed to identify the missed pilgrims. The proposed system is designed based on iris recognition. The iris recognition system is a classical biometric recognition system, and it is considered as one the fastest and accurate recognition systems. Due to overcrowd either in Mecca or Madinah, the police officers will be notified for missed pilgrims. Basically, the missed pilgrims miss their group members for any reasons. The police officers role here is to capture an iris image for the missed pilgrim using their own mobiles. The mobile camera is supported with zoom lenses to have better quality image. Based on the online image capturing, the officer directly uses the Missed Hajj mobile application which is installed on his mobile. The application uses the captured image and search among the huge database at the ministry of Hajj in order to identify the missed pilgrim [2]. The mobile zoom lens is shown in figure 4.
Figure 4: The Mobile zoom Lens |
Feature Extraction
An efficient representation by using a set of numerical values of the input image is the main goal the feature extraction phase. Furthermore, another goal for feature extraction is removing the redundant data from the input images. Later on, thesefeatures are mapped into a chosen classifier in order to identify and verify the input image. In this phase, the normalized grey iris image is used to extract the features based on the Discrete Cosine Transform (DCT).
The DCT coefficients are well known for reducing the redundant data. It focuses on computing the energy for the input image. The DCT features using the DCT coefficients are used in the proposed system. Extracting the DCT features is done firstly by finding the DCT of the input normalized grey iris image using equation (1) [3][5]
After applying DCT to the full normalized grey iris image, the DCT features are extracted in a vector sequence by applying the zigzag order to the DCT image of the normalized grey iris image. Basically, there are two well-known methods to get 1D vector from the 2D DCT coefficients. The zigzag order is illustrated in Figure 5, where Figure 3(a) is used the proposed system.
Figure 5: The Two zigzag order Methods |
The feature matrices normalized into the range of [-1, 1]. Each normalized grey iris image is represented by a feature vector of size 50 which is empirically tested [4][5].
Feature Extraction Algorithm
for k=1 to number of images Img_in=Read the normalized grey iris image Find the 2D DCT for the Img_in Img_dct= DCT2(Img_in) Apply Zigzag order to obtain 1D vector of the Img_dct Choose the first 50 DCT coefficients of the 1D vector |
Classification (Pattern Matching and Verification)
The Wavelet Probabilistic Neural Network (WPNN) classifier is applied to the proposed system as a classifier. The four layers of the WPNN architecture is shown in figure 6.
Figure 6 The Wavelet Probabilistic Neural Network |
The four layers are classified as feature layer, wavelet layer, Gaussian layer and finally the decision layer.
Firstly, the feature layer, X_{1},…,X_{N} are the input data which are the sets of feature vectors, where N is the dimension size of data.
Secondly, in the wavelet layer: this is a linear combination of several multidimensional wavelets. Each wavelet neuron is equal to a multidimensional wavelet. The wavelet is computed using equation 2.
Each wavelet is considered as a family function generated from the single function Φ(x). Both translation and scaling were used in computing the wavelet in order to be localized in time space and the frequency space respectively. The Φ(x) is known as the mother wavelet and both a, b parameters are known as scaling and translation factors respectively[7]. Finally, in Gaussian layer the probability density function of each Gaussian neuron is computed using equation 3.
where X is known as feature vector, p is known as the training set dimension size, n is known as the input data dimension size, j is known as the jth data set, S^{i}_{j} is known as the training set and σ is known as Gaussian function smoothing factor.
The entire scaling factor, the translation factor and the smoothing factor are randomly initialized at the beginning in order to be trained by The Particle Swarm Optimization (PSO). Based on the completion of the training, the WPNN architecture, all the parameters will be fixed for more and further verification.
Learning Algorithm
The PSO was developed by James Kennedy and Russell Eberhart in 1995. The PSO is used for training a single neuron for optimizing the model of the WPNN architecture. Basically, the PSO is considered as a new bio-inspired method in optimization [6]. The main algorithm looks for finding the best solution in the search space iteratively using the particle movement.For example, at any time unit t, the position of ith particle x_{i}, i =1,2,…,M, where M is the number of particles, moves by summing the velocity vector v_{i}. v_{i} is the function of the best position p_{i} found by that particle, and the best position g can be found among all particles of the swarm. Basically, the movement of the particle is formulated using equations 4 and 5 respectively. :
The w(t) is known as the inertia weight, c is known as the acceleration constants, and μÎ(0,1) is known as the uniformly distributed random numbers. Both the wavelet neuron and the Gaussian factors are encoded. The scaling and translation factors are used to encode the wavelet neuron. In addition, the smoothing factor is used to encode the Gaussian factor. The PSO is used to search for the best set of factors in the multidimensional space [7].
Decision
Finally, there are five output probabilistic indicated values for iris features. Those probabilistic indicated values are located in In the last layer of WPNN, the decision layer. By finding the average of those five probabilistic indicated values which are called. In the face features, only one output probabilistic value is computed and it is called. The average value of both probabilistic values the and the is computed. By adjusting the threshold of the average value the false rejection ratios (FRR) and false accept ratios (FAR) are obtained.Figure 7 illustrates both FRR and FAR ratio [7].
Figure 7: The curve of both FRR AND FAR |
Using multimodal biometrics recognition, the decision threshold is shown in the horizontal axis. Each false rate is shown in the vertical axis. From figure6, there are two main points are concluded. They are:
a) When the value of P_{av} for an unregistered sample is less than the decision threshold, a false accepted occurs. The FAR is computed by counting all the false acceptance trails.
b) When the value of P_{av} for an unregistered sample is greater than the decision threshold, the registered sample is wrong rejected. The FRR is computed by counting all the trials of false rejection
Mobile Application
Upon completeion of all the recognition phases, a mobile application is developed to be installed on he smart mobile phones supporting both systems the Andriod and the IOS.
Basically, when the proposed system runs on the smart mobile phones, an iris image will be taken for the pilgrim which will be sent to Ministry of Hajj server for further analysis. Here, the huge database will be accessed and the iris recognition is run using image search engine.
Experimental results
Obtaining a high recognition rate for any recognition system requires a huge database for both trainingg and testing. several experiments are performed on the iris database in order to evaluate the proposed system. The WPNN classifier is used to measure the recognition rate of the proposed system. The recognition rate of the proposed system is 86% which is acceptable.
Due to the fact that the iris last for twenty hours after death, the proposed system works well for identifying the dead pilgrims who passed away less than twenty four hours; however, the proposed system does not work for the dead pilgrims one day or more from their death.
Conclusion
A new framework recognition system for identifying the miss pilgrims in both Hajj and Umrah is proposed in this paper, The WPNN classifier is used in the proposed system based on DCT features. The features are extracted using the DCT coefficients. Finally, a mobile application for both Android and IOS system is proposed as well. The results of the proposed system and its performance based on WPNN classifier are very good comparing to the existing ones.
Acknowledgment
I would like to express my sincere appreciation and thanks owe thanks to TAIBAH University, College of Computer Science and Engineering (CCSE), Kingdom of Saudi Arabia for supporting this research.
References
The spectrum is a precious electromagnetic resource and is regulated by governmental agencies in order to manage complex issues. Presently, spectrum is allocated by fixed allocation policy in which transmission power is regulated and different frequency bands are assigned for different services and applications. There has been tremendous growth of wireless users and applications over the last decade. The total number of users worldwide were 3.2 billion in 2009 and it was projected to increase by 100 folds by 2013^{1}. These new applications require more spectrum allocation and it becomes difficult to find unallocated spectrum as most of the spectrum bands stands already allocated by fixed allocation policy. This policy has created a situation where there appears an artificial scarcity of the spectrum. But the survey of Federal Communications Commission (FCC)^{2} spectrum task force reported the utilization of the allocated spectrum varies from 15% to 85% and is a function of space and time. Other spectrum occupancy measurements conducted have revealed that the average utilization in New York City^{3} was at 5.2% while in Chicago^{4} at 17.4%. Several other countries spectrum occupancy measurement studies^{5-10} also confirmed that spectrum is heavily under-utilized at this moment. Thus, the persistent increase in demand of spectrum cannot be fulfilled unless an alternate scheme to regulate the scarce spectrum in not found. This new scheme is the dynamic spectrum access (DSA)^{11-13} wherein cognitive (or secondary) users are allowed to opportunistically utilize the idle licensed bands, referred to as spectrum hole or white space, without interfering with the existing (or primary) users. If spectrum is utilized on time or frequency basis, the spectrum opportunities appear in the form of holes. In time domain, it is the period of time during which the primary user in not transmitting and in frequency domain, it is the frequency band in which secondary user can transmit without interference to primary user. The enabling technology of the DSA is the cognitive radio (CR). According to Haykin^{14}, the cognitive radio is defined as an intelligent wireless communication system that is aware of its surrounding environment (i.e., outside world), and uses the methodology of understanding-by-building to learn from the environment and adapt its internal states to statistical variations in the incoming RF stimuli by making corresponding changes in certain operating parameters (e.g., transmit-power, carrier frequency, and modulation strategy) in real-time, with two primary objectives in mind:
Cognitive radio has two important capabilities: cognitive capability and reconfiguration. Cognitive capability enables the CR to sense its radio environment in order to find spectrum holes. These spectrum holes will be utilized to communicate wherever and whenever needed, thus increasing spectrum utilization.
Reconfigurability is the capability of adjusting operating parameters for the transmission on the fly without any modifications in the hardware components and thus, enable the cognitive radio to adapt easily to the dynamic radio environment. The reconfigurable parameters are operating frequency, modulation and transmission power etc.
Functions of Cognitive Radio
The main purpose of the cognitive radio is to realize dynamic access to the idle spectrum in order to have communication. However, implementation of a cognitive radio is a challenging task. A typical cognitive radio scenario consists of secondary users (SUs) which co-exist with the primary users (PUs). PU has a priority to use the spectrum as they have legacy rights for the spectrum access. SU opportunistically access the spectrum when they find primary user is not using the spectrum or underlay access to the spectrum, when both PU and SU co-exist, with strict constraint over non-interference among the users. The development of cognitive radio technology is still at an early stage due to multitude of challenges and these issues need to be addressed rigorously before a practical cognitive radio network is realized. The main functions of cognitive radio are classified as under^{13}:
Spectrum Sensing
Spectrum sensing is the important function of the cognitive radio. The cognitive user (or SU) has to sense the radio environment in order to detect the idle spectrum (or spectrum hole) and use that idle spectrum for its communication. The sensing operation should also be able to detect the PU’s arrival on the frequency band, presently occupied by SUs, instantly and accurately^{15}.
Spectrum Management
Spectrum management helps in acquiring the best spectrum hole for the transmission of SU among large number of available spectrum holes. The SU has to interact with different protocol layers of the network to maintain the QoS of an application. Based on the availability of the spectrum, the channel is allocated to the SU. This decision also depends on the internal and external policies as well as other regulatory issues of the spectrum. The spectrum management functions can be classified as under:
Spectrum Sharing
Spectrum sharing is an important functionality of cognitive radio as it coordinates the traffic between secondary and primary users. It is a challenging task as it requires high degree of cooperation, understanding and coordination between primary and secondary users.
Spectrum Mobility
Spectrum mobility refers to the changing the frequency band during data transmission due to arrival of PU on that band. The SU has to switch to another frequency band which is not used by the PU at that time in order to continue the transmission^{16}. The switching to this idle band should be seamless so that there is minimum QoS degradation of the application running on the SU. As soon as the PU needs the frequency band, the SU has to terminate its transmission and free the frequency band for the PU functioning^{17}.
Handoff Requirements
The temporary service disruption occurs during handoff which will influence the QoS of the communication. The necessary requirements to reduce the adverse effects of handoff are^{18}:
Handoff Issues in Cognitive Radio Networks
In addition to the issues related to handoff in other wireless networks, the cognitive radio has the new challenge of spectrum mobility. Therefore, the solutions available in the literature for other wireless networks cannot be applied to cognitive radio due to absence of fixed spectrum allocation. The other challenge is due to the availability of the very wide range of spectrum for transmission. The channel characteristics change with change in operating frequency that depends on the idle band found. In this section, we will provide the survey of previous research works for important function of spectrum handoff.
The important issue of modeling the spectrum handoff in presence of multiple interruptions due to arrival of primary users was analyzed in^{17,19-21}. For the analysis of spectrum handoff, two approaches were adopted in research literature depending on the timing of the selection of the channels to be used at the time of handoff. The first approach is the proactive-decision approach^{22-25} in which the channels to be used for future handoff is decided before actual handoff. The channel selection is based on the use of prediction techniques. The second approach is the reactive decision approach^{26-28} in which the channels are selected after handoff request is made through instantaneous sensing of the spectrum. Comparative study of the two approaches is provided in^{29} which discussed their advantages and disadvantages. In^{30} hybrid spectrum handoff algorithm is proposed in which the algorithm switches between proactive and reactive approaches depending upon the primary arrival rate with the aim to reduce the service time of the SU. The authors of ^{31} presented the analysis of handoff for opportunistic and negotiated situations in terms of four metrics such as link maintenance probability, number of handoffs, non-completion probability and switching delay. The authors of ^{32-33} proposed characterization of PUs and SUs by assuming an exponential traffic distribution model. The authors of^{32} analyzed the impact of channel reservation on handoff while^{34} studied the forced termination probability, blocking probability and throughput by assuming fixed, truncated exponential, truncated lognormal and truncated pareto traffic distribution models. The authors of ^{35} proposed a metric such as overall system time with the aim to minimize it in order to support better QoS of secondary communication. The concept of spectrum pooling was proposed to cognitive radio^{36} in order to realize a real time handoff while^{37} proposed the division of spectrum pool into two parts i.e. inside and outside bands. When inside bands reduce below a threshold value, the outside band is sensed to find idle spectrum and^{38} proposed the concept of second receiver in addition to spectrum pool in order to support real time handoff. The authors of ^{39} proposed combined optimization of spectrum sensing and handoff in order to obtain improvements in both functions and realize a fast handoff while^{40} proposed the cooperative sensing of secondary user group and admission control for multiuser scenario. By adopting that policy there was an increase in accuracy of PU’s detection probability, reduction in probability of missed detection and false alarm thus, resulted in increased efficiency of the spectrum handoff. The authors of ^{41} proposed a voluntary handoff in which the SUs handoff voluntarily to another idle channel before PU’s detection on that channel in order to reduce disturbance during handoff and at the same time minimize handoff delay which reduced to switching delay only while^{42} proposed selection of target channel sequence decided proactively so that the handoff failure rate is reduced. The authors’ of ^{43} proposed an algorithm based on time estimation which determined the remaining idle period of the channels and schedules channel usage in advance. The proposed approach reduced the disruption to PUs and at the same time increased channel utilization. The authors’ of ^{44} considered the practical limitation of sensing reliability and sensing time and determined the optimal channels by utilizing partially observable markov decision process to reveal the network information by partially sensing the available channels without the necessity of obtaining correct channel information. The proposed algorithm selected the optimal channels for handoff with minimum waiting time while^{45} proposed an algorithm for sensing and selecting channels having maximum probability of appearing idle using traffic prediction technique that resulted in reduction of the corresponding sensing time. The authors’ of ^{46} proposed to switch from overlay to underlay mode by reducing the transmission power upon PU’s arrival under the non-interference constraint and proposed a multi-cell spectrum handoff in order to overcome the coverage issue of underlay mode while^{47} proposed compromise decision either to stay or change channels during handoff depending on the delay bound requirement of the flow. The proposed algorithm used the cumulative probability based on past backlog measurements in order to take the handoff decision.
Fuzzy logic was applied to handoff ^{48-49} which did two important functions, the power modification and intelligent handoff decision based on the information of interference, transmission power and required data-rate. In ^{50} , the new parameter such as holding time of the channel is included into decision making so that the channel having larger idle period among the available idle channels is selected for transmission after handoff. As a result, there is considerable reduction in handoff rate. There are few works on cognitive adhoc networks that utilize spectrum handoff. The authors’ of ^{51} studied the impact of user mobility and PU’s arrival on the spectrum handoff rate and session continuity distribution. The works^{52-54} applied proactive decision approach for handoff in cognitive adhoc networks and^{55} provided the characterization of spectrum handoff using three dimensional discrete time markov chain and analyzed the effect of different channel selection schemes on the performance of handoff.
Conclusion
Spectrum is a valuable natural resource in this information age. Considering the difference between spectrum efficiency and utilization, novel spectrum management techniques are under study for addressing future needs. The promising solution is the cognitive radio. However, spectrum mobility and handoff are new challenges for CR networks. The spectrum handoff is an important operation to support resilient and continuous communication. In this paper, we surveyed a number of such challenges and presented number of solutions with main focus on spectrum handoff. These issues need to be addressed before a practical cognitive radio network is realized.
References