The paper analyzes two models of mixed integer program- ming and one linear programming model for optimization of the OWA criterion. Experiments were conducted to compare the computational eﬃciency of diﬀerent formu- lations of these models. Based on the obtained results it can be concluded that the redundant constraints added to MILPmodels of OWA can signiﬁcantly shorten the compu- tational time for certain types of localization problems (cer- tain classes of OWA weights vectors). Secondly, the model M1 appears to be much more eﬃcient than the model M2. Besides, if the problem has special structure, which allows one to formulate OWA criterion as standard linear formu- lation, this should be exploited, as it greatly increases its computational eﬃciency. However, adding the redundant constraints to the linear programming OWA formulation does not help and may increase the computational time. Because the results presented here are based on an average solution time, it seems desirable to conduct a more de- tailed statistical analysis (e.g., minimum, maximum, vari- ance) of the results. Perhaps it will allow to ﬁnd new depen- dencies and determine more detailed model characteristics. Better eﬃciency of the model M1 suggests also an oppor- tunity to apply it to quadratic assignment problem (QAP), from which some transformations for the model M2 have been exploited .
 E. Taillard, “Benchmarks for basic scheduling problems”, Eur. J. Oper. Res. , vol. 64, pp. 278–285, 1993.
Jarosław Hurkała received his M.Sc. degree in Computer Sci- ence with honors from the War- saw University of Technology, Poland, in 2010. Currently, he is a Ph.D. student in the In- stitute of Control and Compu- tation Engineering at the War- saw University of Technology. His research area focuses on scheduling problems, heuristic algorithms, fairness and multicriteria optimization. E-mail: email@example.com
 Y. Li, L. Cimini, and N. Sollenberger, “Robust channel estimation for OFDM systems with rapid dispersive fading channels”, IEEE
Trans. Commun ., vol. 46, no. 7, pp. 902–915, 1998.
 A. Youssefi and J. El abbadi, “Pilot design optimization using mod- ified differential evolution algorithm in SISO and MIMO OFDM systems”, J. Basic Appl. Scient. Res., part V, 2012.
in band 1800 MHz devoted to the development of LTE system and thus of the mobile internet infrastructure in Poland . The winners of the competition were: operator Play with 3 blocks of 5 MHz each (total 15 MHz) and T-Mobile with 2 blocks (10 MHz). As a result of this tender the State Treasury received around a 950 mln PLN income. It indicates a high interest in the LTE development not only Solorz’s holding, but also other operators. Another tender (as an auction) is expected in 2013 for that part of band 800 MHz , which is the digital dividend linked with the digital switchover of the terrestrial TV in Poland and was occupied for the military applications until 2012. The spectrum of 2.6 GHz (and higher bands) may also be available, however, the band 800 MHz is mostly attractive due to a relatively lower network invest- ment cost than in the case of higher frequencies. UKE pres- ident Mrs. Magdalena Gaj has estimated  that due to making this frequency spectrum  available to investors, LTE can encompass 90% of the country’s territory, and the planned auction for band 800 MHz frequencies will allow to extend LTE services to less developed areas.
There is currently no better communication technology available than IP. It is extremely flexible and has been implemented worldwide. The modern Internet of Things (IoT) concept is also based on IP technology. The number of applications is growing rapidly, with video applications especially on the increase recently. It is common knowl- edge that video services are extremely resource consum- ing. So, congestion will occur rapidly in certain segments of the Internet. Packet losses, end-to-end delays and jitter are the inevitable consequences. They will be reflected by a rapid deterioration in service quality, and customer sat- isfaction will plummet. New techniques will have to be adopted to address the situation. One such new technique is the MPEG-DASH method (Dynamic Adaptive Stream- ing over HTTP), that was standardized in 2012 as ISO/IEC 23009-1:2012 . What effects does it have on QoS in IP networks? The work described in this paper aimed to find an answer to that question.
increasing its computational complexity. Therefore, the ef- ficient formulations for OWAoptimization are sought. We introduce general MILPmodels that can be applied for any non-negative preference weights. It extends the LP for- mulation, adding the mixed integer part. We also propose some simple valid inequalities to improve the computational performance. The results show the advantage of proposed new hybrid formulations over other general MILPmodels from literature for some specific problem types. The great- est improvement is obtained for trimmed mean problems. This is particularly important as trimmed mean problems seems to be one of the most useful in practical applications from all other problem types with non-monotonic prefer- ence weights, which can not be solved by LP model. On the other hand, hybrid models perform very poorly for prob- lem types T2–T5. However, these types represent rather artificial objective functions (preferences) with little practi- cal value. The proposed models perform surprisingly well for problems with increasing weights, which require the largest number of binary variables. For example, consid- ering problems T6, the hybrid models obtain much shorter solution times than previous general formulations.
Q ( α ) d α . (15) Thus, the orness of a RIM quantiﬁer is equal to the area under it. The measure takes the values between 0 (achieved for Q(1) = 1 and Q( α ) = 0 for all other α ) and 1 (achieved for Q(0) = 1 and Q( α ) = 0 for all other α ). In particu- lar, orness(Q) = 1/2 for Q( α ) = α which is generated by equal weights w k = 1/n. Formula (15) allows one to de- ﬁne the orness of the WOWA aggregation (4) which can be viewed with the RIM quantiﬁer Q( α ) = w ∗ ( α ) . Let us consider piecewise linear function Q = w ∗ deﬁned by weights vectors w of dimension n according to the step- wise generation function (8). One may easily notice that decreasing weights w 1 ≥ w 2 ≥ . . . ≥ w n generate a strictly increasing concave curve Q( α ) ≥ α thus guaranteeing the or-likeness of the WOWA operator. Similarly, increasing weights w 1 ≤ w 2 ≤ . . . ≤ w n generate a strictly increasing convex curve Q( α ) ≤ α thus guaranteeing the and-likeness of the WOWA operator. Actually, the monotonic weights generate the totally or-like and and-like operators, respec- tively, in the sense that
While significant decision models are being presented in these papers, but, very few studies have considered all cri- teria relevant to rural telecommunications, and most of them obviously apply no factor interactions. For exam- ple, if a model’s emphasis is mainly technical, then the economic, social, regulatory and environmental criteria are probably not adequately addressed. Basically, the AHP is a suitable method when optimization is not pursued, re- sources are not restricted, and interdependencies between factors do not exist . However, such models do not consider important issues such as interaction among and between decision making levels/clusters as well as depen- dency among qualitative factors. These are important issues in rural telecommunications decision problems which can- not not be structured hierarchically because they involve many interactions and dependencies requiring a MCDM method to holistically deal with qualitative and quantita- tive data, with different conflicting objectives, to arrive at a consensus decision in relation to the choice of a suitable rural telecommunication technology.
A speech signal consists of different attributes, such as loudness, voiced/unvoiced sounds, pitch, fundamental fre- quency, spectral envelope, formants etc. These attributes help identify the speaker and speech features . Although speech recognition and speaker recognition are different fields, the feature extraction methods in both fields over- lap . These methods include predictive models based on the linear predictive coding coefficient (LPCC), percep- tual linear prediction (PLP), mel frequency cepstral co- efficient (MFCC) and relative spectra filtering (RASTA). These methods can be implemented in speech recognition as well as in speaker recognition –. Speech fea- tures can be optimized for improving recognition accuracy with the help of various optimization algorithms, like the genetic algorithm (GA), particle swarm optimization, ant colony search algorithm, etc. . GA can be used, in deep neural networks, for improvement in recognition ac- curacy –. In past studies, many researchers have implemented GA with an artificial neural network (ANN), i.e. Lan et al. . They have implemented GA, instead of the steepest descent method, for updating weights and achieved a 91% recognition accuracy. Balochian et al.  claimed a 96.49% accuracy level by using GA with the multi-layer perceptron (MLP) classifier.
In all the methods discussed in Section 3, information about the local communities are not considered while calculating similarity index. By considering local information about community while calculating similarity index, precision of link prediction increases as shown by Soundarajan and Hopcraft  for Common Neighbor and Resource Alloca- tion. But the method proposed by them is not being further explored over other link prediction methods and other bench mark datasets. It will be interesting to investigate similarity based link prediction with local information about commu- nity as it can be further extended to Louvain method  proposed by Blondel et al.
systems are available, because they contain the phase in- formation of the underlying linear system in contrast to second order statistics, and they are of great value in ap- plications, such as radar, sonar, array processing, blind equalization, time delay estimation, data communication, image and speech processing and seismology –. Many algorithms have been proposed in the literature for the identification of FIR system using cumulants. These algorithms can be classified into three broad classes of solutions: closed form solutions , , , opti- mization based solutions ,  and linear algebra so- lutions –. The linear algebra solutions have re- ceived great attention because they have “simpler” compu- tation and are free of the problems of local extreme values that often occur in the optimization solution. Although, the closed-form solutions have similar features, they usu- ally do not smooth out the noises caused from the obser- vation and computation. Therefore, while these solutions are interesting from the theoretical point of view, they are not recommended for practical applications , . The main goal of this investigation is to elaborate an accurate and efficient algorithm able to estimate the moving aver- age (MA) (or FIR) parameters in noisy environment. So, we address the problem of estimating the parameters of a FIR system from the output observation when the system is excited by an unobservable independent identically dis- tributed (i.i.d.) sequence. The proposed algorithms, based on third and fourth order cumulants, to estimate the parame- ters of MA process when the order is known, are presented. For validation purpose these method are used to search for a model able to describe the broadband radio access net- work (BRAN A and BRAN E) channels, represented by a FIR model.
We can notice that this QoS support does not need any additional message. The Hello and TC messages of OLSR are extended with QoS information in order to allow any flow source to compute the shortest route providing the bandwidth requested by its new flow. As the problem of finding a route meeting a given bandwidth has been shown NP-hard in wireless networks subject to radio in- terferences , we use an approximation to compute the bandwidth consumed by a flow at the MAC level. This approximation is used only by the QoS routing protocol to select the route which also depends on the local avail- able bandwidth measured at each node. Once a route has been found for a QoS flow, it is used by all packets of the flow considered, until either a shorter route is established because network resources have been released, or it is no longer valid because of a link breakage. Source routing can be used for that purpose. Notice that BE flows are routed hop-by-hop.
Abstract—The paper discusses quality of service in LTE and LTE-A networks seen as a challenge that can be met with Self- Organizing Networks (SON) functionalities. The SON con- cepts have been included in the LTE (E-UTRAN) standards since the first release of the LTE technology. Self-optimization functionalities will monitor and analyze performance mea- surements, notifications, and self-test results and will automat- ically trigger re-configuration actions on the affected network nodes when necessary. The SON specifications have been built over the existing 3GPP network management architecture, the ultimate implementation of SON in 4G networks will bring many advantages. Successive SON procedures are waiting for their time and money to be implemented in 4G, though some essential issues for example of inter Radio Access Technology (RAT) interfaces must be overworked.
the following experiments. For the Gaussian kernel, pa- rameter tuning was reduced to choice of the parameter σ , which determines a range of training set points inﬂuence. In case of polynomial kernels, the tuning concerned poly- nomial order (parameter m of equations (11) and (12)). Three diﬀerent values for m were tested throughout exper- iments: m = 2 , 3 and 6. One needs to note, that due to high dimension of the original data vectors (d = 25 ), even for the considered low polynomial orders, a resulting fea- ture space, where classiﬁcation gets performed, has very high dimensionality. As it was shown in , cardinality of the H space, in case the polynomial (11) is considered, is related to a polynomial order m and to an original input vector dimension d via the formula:
Abstract—Numerous practical engineering applications can be formulated as non-convex, non-smooth, multi-modal and ill-conditioned optimization problems. Classical, determin- istic algorithms require an enormous computational effort, which tends to fail as the problem size and its complexity in- crease, which is often the case. On the other hand, stochastic, biologically-inspired techniques, designed for global optimum calculation, frequently prove successful when applied to real life computational problems. While the area of bio-inspired algorithms (BIAs) is still relatively young, it is undergoing con- tinuous, rapid development. Selection and tuning of the ap- propriate optimization solver for a particular task can be chal- lenging and requires expert knowledge of the methods to be considered. Comparing the performance of viable candidates against a defined test bed environment can help in solving such dilemmas. This paper presents the benchmark results of two biologically inspired algorithms: covariance matrix adapta- tion evolution strategy (CMA-ES) and two variants of particle swarm optimization (PSO). COCO (COmparing Continuous Optimizers) – a platform for systematic and sound compar- isons of real-parameter global optimization solvers was used to evaluate the performance of CMA-ES and PSO methods. Particular attention was paid to the efficiency and scalability of both techniques.
5. Cross Layer Approach
To improve network performance, interaction of parameters across the protocol stack is necessary. Energy is a param- eter of the physical layer and routing is considered at the network layer. Layers need to interact to obtain the value of energy in a routing packet. This helps the routing protocol to choose an energy efficient path. Route energy packets which are used to exchange energy values among nodes are generated using the cross-layer design. Hoesel et al. presents a cross-layer approach in which the routing proto- col uses topology and infrastructure information available at the MAC layer . It reestablishes the route utiliz- ing information at the MAC layer and outperforms S-MAC and Dynamic Source Routing (DSR) in mobile sensor net- works . Cross Layer MAC (CL-MAC) makes and op- timizes scheduling decisions based on cross layer informa- tion . Path-Loss Ordered Slotted Aloha (PLOSA) pro- tocol is designed using cross-layer design for wireless data collection networks . It helps in observing physical sig- nals and orders the access of nodes accordingly. Nodes at a greater distance from the collector get an earlier chance to access the slot of the transmission channel. PLOSA has a high delivery rate and low latency.
Fig. 2. Process of analysis of decision alternatives for proxy cache configuration.
3. Decision support tool concept In order to meet requirements presented earlier the tool sup- porting web cache management should be able to perform several tasks. These tasks should be organized in a process that leads to reasonable choice about better configuration for examined cache. Figure 2 presents such process. In this process, for a given representation of stream of doc- uments requests a number of possible sets of proxy cache parameters are examined. Each set of parameters (config- uration) can be seen as a decision alternative. Collected performance results are used to evaluate these alternatives. Multicriteria analysis methodology based on mathematical modeling is used for the purpose of analyzing decision al- ternatives. The effect of such analysis is the best – accord- ing to preferences of person performing analysis – configu-
2. Related Works
Some research has been introduced to classify data by using distance to the centroid. This includes , , –, which have used that distance to generate features. The basic idea of those methods is similar, that is, the distance between a point and the respective centroid is used to dif- ferentiate data within dataset. The differences between that researches are how the centroid is extracted and how that distances is used. Some methods use k-means by using the previously defined number of clusters , ,  and by partitioning the dataset based on their label and by extract- ing the centroids according to their average features in each partition , , .
since offered PU traffic affects the PU users directly and to SU indirectly.
In this paper the performance of a cognitive radio network using M/M/n/k traffic model based on 2-D Markov Chain is analyzed. The results shows the performance of the net- work varying different traffic parameters, provide expected results. Still the scope to use M/G/1/K traffic of packet switch network of finite queue has to be researched. In this case two dimensional traffic models using state tran- sition chain will not be possible because of general PDF service time. The authors can apply tabular form of 2-D traffic model of M/G/1/K to pave the way for evaluating packet loss PU and SU probability. Finally the impact of fading channel on false alarm and misdetection can also be included on the traffic model to observe the performance under small scale fading environment. One of the major components of 5G mobile communications is the concept
TSD2 comprises of 100 points, each indicating the num- ber of packets transmitted. The forecast horizon is taken as 10%, which implies 10 data points, and one-step ahead pre- dictions are obtained for these points. A three-step ahead prediction is carried out by using a forecast horizon of 12 which is again nearly 10%. The MAE and MSE perfor- mance results for all the models in both these cases are presented in Table 2. The original TS is shown in Fig. 9. The predictions for the one-step ahead forecast and three- step ahead forecast are shown in Figs. 10 and 11 respec- tively. It is noticed that the MA-filter based hybrid ARIMA- ANN model outperformed the others in terms of both MAE and MSE.