Abstract

When the abstract was writing this master's work was not complete yet. Final completion: june 2020. Full text of work and materials on the topic can be obtained from the website after this date.

Contents

• Introduction
• 1. Relevance of a subject
• 2. Purpose and objectives of the study, planned results
• 3. Research and Development Overview
• 3.1 Analysis of wireless networks
• 3.2 Analysis of neural networks
• 3.2.1 Basic artificial model
• 3.2.2 Application of neural networks in telecommunication systems
• 4. Wireless Network Management Algorithm Preview
• Conclusion
• References

Introduction

In places of great accumulation of people, such as stadium, city centers, arenas, conference–centers, train stations, mobile networks are especially overloaded. The growth of data traffic is ensured by smartphones and mobile applications of laptops and tablets. Therefore, mobile operators need tools to address network bandwidth. Build Wi–Fi networks on high–density objects & ndash; relevant task today.

1. Relevance of a subject

The relevance of the work is that servicing of various infotelecommunication networks is quite expensive and energy–intensive meropyrization. Therefore, the method of servicing the Wi–Fi wireless network will be considered in order to reduce the energy consumption on the maintenance of the network so as to reduce the cost of power supply of the wireless network equipment.

Masterʼs work is devoted to the current scientific task of developing an algorithm for solving problems of equipment operation control in wireless networks, using the results of analysis of the current load of the network and prediction of change of characteristics of the provided quality of service in the network depending on the parameters of the received request for data transmission.

The results obtained can be used in the design of high–speed wireless networks with a high concentration of subscribers with economical irrigation.

2. Purpose and objectives of the study, planned results

The aim of scientific work is to develop an algorithm for optimal management of wireless network resources under conditions of high concentration of subscribers.

When developing such an algorithm, it is necessary to solve the following problems:

1. analyse the methods of a possible solution;
2. develop network maintenance requirements;
3. construct a mathematical model of the wireless network resource management algorithm;
4. evaluate performance – calculation of influence or determination of signal–to–noise ratio;
5. to check operability of network.

3. Research and Development Overview

technology originated from a decision by the United States Federal Communications Commission (FCC, 1985) to open multiple bands of wireless spectrum for use without a government license. These strips have already been used for all kinds of equipment such as microwave ovens. To operate in these frequencies, devices must use spectrum propagation technology. Due to this technology, the radio signal propagates over a wider frequency range, making the signal less sensitive to interference and difficult to intercept.

3.1 Analysis of wireless networks

There are many objects with a high concentration of subscribers where wireless networks cannot be avoided. In most cases, wireless networks offer certain advantages over wired networks:

lack of wires;

minimum of installation and construction works;

high speed;

cheap installation and ownership;

flexibility in construction;

reconfiguration and scalability;

reconfiguration and scalability;

In everyday life in places of large accumulation of people, a type of wireless network has gained greater use: Wi–Fi based on IEEE 802.11 technology [1]. Of all the standards of such IEEE 802.11 data technology, four are most commonly used in practice: 802.11a, 802.11b, 802.11g and 802.11n.

In the IEEE 802.11b standard, the data rate is up to 11 Mbps, running in the 2.4 GHz band, a standard that has gained the most popularity among wireless network equipment manufacturers. Since equipment operating at a maximum speed of 11 Mbps has a shorter range than at lower speeds, 802.11b provides for automatic speed reduction when signal quality is degraded.

The IEEE 802.11g standard is a logical development of 802.11b and involves data transmission in the same frequency band. In addition, 802.11g is fully compatible with 802.11b, meaning that any 802.11g device must support 802.11b devices. The maximum transmission rate in 802.11g standard is 54 Mbps, so today it is the most promising standard of wireless communication.

802.11n increases data transfer speeds almost fourfold compared to 802.11g devices (with a maximum speed of 54 Mbits/s) when used in 802.11n mode with other 802.11n devices. In theory, 802.11n can provide data rates up to 480 Mbps. 802.11n devices run in the 2.4 – 2.5, or 5.0 GHz bands.

Wireless network problems always remain:

limitation of range of communication;

a sharp drop in network capacity with an increase in the number of subscribers.

he mismatch between the planned bandwidth of the wireless LAN and the rapidly growing traffic of its users leads to a significant deterioration of the network performance, dissatisfaction of its users and incorrect conclusions that the Wi–Fi network cannot cope with the heavy load. However, following simple design principles will ensure sufficient Wi–Fi bandwidth to serve thousands of users in one place, as examples of successful projects show.

Let 's take a look at the basic requirements for the wireless network being designed:

the type of data transmitted (data transmission, voice or positioning);

density of users;

requirements to a covering;

features of client devices (transmitter power, supported ranges and channels, supported data rates);

network security requirements.

Depending on the type of data to be transmitted over the network, the average link capacity and transmitter range to be used in the network are required. The average bandwidth of a data link depends on quality of service requirements and is difficult to achieve.

The necessary and sufficient number of subscribers that can connect to one access point requires transmitter power and frequency channel allocation. The lower the transmitter power, the lower the speed of the organized channel and the less it is possible to connect subscribers to one access point. Optimal number of subscribers – from 13 to 18.

The security of wireless networks depends heavily on the use of a number of different technologies: encryption, digital signature, passwords, etc. How these technologies are used has a strong impact on network security. The security of wireless networks is subject to a large number of studies and this study does not address it.

A major problem in designing or expanding wireless access networks is not overlapping the frequencies of neighboring access points to avoid interference and reduce transmission rates. This is usually done by setting up adjacent points on non–overlapping channels.

As a rule, to properly plan the site of access points, first examine the functional dependencies of the following values:

range;

used channels of the selected range;

transmitter power;

antenna type and gain;

allowed data rates.

3.2 Analysis of neural networks

In the last few years we have seen an explosion of interest in neural networks, which are successfully applied in a variety of areas & ndash; Business, medicine, engineering, geology, physics. Neural networks have come into practice wherever it is necessary to solve problems of prediction, classification or control. There are several reasons for this impressive success [3]:

It is a lot of opportunities::
Neural networks & ndash; An extremely powerful simulation method that reproduces extremely complex dependencies. In particular, neural networks are nonlinear in nature. Over the years, linear modeling has been the primary modeling technique in most areas because optimization procedures are well developed for it. In tasks where linear approximation is unsatisfactory (and there are quite a lot of such), linear models do not work well. In addition, neural networks cope with the curse of dimension, which does not allow modeling linear dependencies in the case of a large number of variables.

Usability:
Neural networks learn from examples. The neural network user picks up representative data and then runs a learning algorithm that automatically perceives the data structure. The user is, of course, required to have a set of heuristic knowledge of how to select and prepare data, select the desired network architecture, and interpret the results, but the level of knowledge required to successfully use neural networks is much more modest than, for example, traditional statistical methods.

Neural networks are attractive from an intuitive point of view, for they are based on a primitive biological model of nervous systems. In the future, the development of such neuro–biological models can lead to the creation of truly thinking computers. Meanwhile, the already simple neural networks that ST Neural Networks is building are a powerful weapon in the arsenal of a specialist in applied statistics.

3.2.1 Basic artificial model

To reflect the essence of biological neural systems, the definition of an artificial neuron is given as follows:

It receives inputs (source data or outputs of other neurons of the neural network) through several input channels. Each input signal passes through a compound having a certain intensity (or weight); This weight corresponds to the synaptic activity of the biological neuron. A certain threshold value is associated with each neuron. The weighted sum of inputs is calculated, the threshold value is subtracted from it, and the result is the amount of neuron activation (it is also called post–synaptic neuron potential – PSP).

The activation signal is converted by the activation function (or transfer function) and the neuron output is obtained.

If you use a step–by–step activation function (i.e., the neuron output is zero if the input is negative, and one if the input is zero or positive), then such a neuron will work exactly like the natural neuron described above (subtract the threshold value from the weighted sum and compare the result to zero – It is the same as comparing a weighted sum to a threshold). In reality, as we will soon see, threshold functions are rarely used in artificial neural networks. Note that weights can be negative, – This means that the synapse does not have an excitatory effect on the neuron, but a inhibitory effect (inhibitory neurons are present in the brain).

It was a description of an individual neuron. Now the question arises: how to connect neurons to each other? If the network is supposed to be used for something, it must have inputs (taking the values of the variables of interest from the outside world) and outputs (forecasts or control signals). Inputs and outputs correspond to sensory and motor nerves – for example, from the eyes and into the arms, respectively. In addition, however, there may still be many intermediate neurons performing internal functions in the network. Input, hidden, and output neurons must be linked together.

The key question here is – feedback. The simplest network has a direct signal transmission structure: Signals pass from inputs through hidden elements and eventually come to output elements. Such a structure has a sustainable behavior. If the network is recurrent (i.e. contains connections leading backward from more distant to more near neurons), it can be unstable and have very complex behavior dynamics. Recurrent networks are of great interest to researchers in the field of neural networks, but in solving practical problems, at least until now, direct transmission structures have been most useful, and this type of neural networks is modeled in the ST Neural Networks package.

A typical example of a direct signal network is shown in Figure 2. Neurons are regularly organized into layers. The input layer is simply used to enter input variable values. Each of the hidden and output neurons is connected to all elements of the previous layer. Networks in which neurons are linked to only some of the neurons of the previous layer could be considered; However, for most full network applications, this is the type of network implemented in ST Neural Networks.

3.2.2 Application of neural networks in telecommunication systems

For a long time it was considered that neurocomputers are effective and applicable only for solving so–called informalizable and poorly formalized problems connected with the need to include in the algorithm of their solving data learning on real experimental material. First of all, such tasks include tasks of image distribution. Recently, the field of application of neuroinformation technologies has been dynamically expanding [3]. They are increasingly used in tasks with pronounced natural parallelism: processing signals, images, etc.

Among the main areas of application of neurocomputators in communication systems are identified [4]:

management of switching;

routing;

traffic control;

channel allocation in mobile radio communication systems.

Solving almost any problem in mobile radio communication systems.

Solving almost any problem in a neural network logical basis implies the following stages [5]:

Generation of input and output signals of NN;

Generation of the desired output signal NN;

Generation of error signal and optimization functionality;

Formation of the NN structure adequate to the selected task;

Development of an algorithm for adjustment of the NN equivalent to the process of solving the problem in a neural network logical basis;

Conducting studies of the problem solving process.

The training method and neural network used to control the operation of a high–quality packet switching network in asynchronous mode are described in [6]. The network is used to control packet switching in the transmission of voice, images and data. The switch is represented as a logic device whose input is N signals and which outputs these signals in any order shown.

Work [8] consider a spatial switch (the inputs and outputs of the switch are different physical lines). Examples of the structure of the NN for controlling the switching process in various telecommunication systems are given.

Currently, the number of connections emulated in a neurocomputer can reach several hundred million. Therefore, it becomes possible to build switches with neural network management on several cells of channels [4]. Work [8] considers Bianca multi–stage switching circuit and its control with the help of neurocontroller. For more information on switching control using artificial NN, refer to [4, 9].

The use of HC for traffic control in complex multi–stage communication systems is proposed in Work [11]. The difficulty of the task is due to the fact that, in the first, parameters characterizing information flows are unknown in advance, and in the second, quality requirements can change over time. NN solves optimization problems related to finding conflict–free flows at specified input and output values. At the same time, the NN easily adapts to changes in conditions.

Setting and solving problems in channel allocation in mobile radio communication systems in a neural network basis differ little from setting and solving a routing task. The difference is in the cellular structure of the radio network and a large number of switched nodes [4].

4. Wireless Network Management Algorithm Preview

Main objective – This is to improve the management of wireless data networks, in our case with Wi–Fi technology, using intelligent methods [10].

The mathematical apparatus chosen for research is the neural network, due to its possible application in prediction of the result. The ability of such a network to predict follows from the ability to generalize and highlight invisible dependencies between input and output data that a person may have missed or omitted due to the difficulty of solving the problem. After training, the neural network is able to anticipate the future meaning of a certain sequence based on several previous values and/or some current factors. It should also be remembered that forecasting is possible only when previous changes to some extent predeterminate future changes. Let's consider possible model. To begin with, select what inputs will be and what outputs will be for these input variables. For the prediction process, statistics are collected on the number of receiving devices located within the area of one transmitter and the signal power levels, statistical parameters of traffic (such as data, voice, video, volume of transmitted data, etc.). This process does not require the intervention of additional equipment connected to mobile devices, since the signal power levels can be measured by means of built–in wireless communication devices. It is also necessary to consider the possible interference that devices create for each other. Based on the collected data, the neural network training process follows, which should predict the power levels of objects and possible optimal settings of access points in the network.

The process of network reconfiguration occurs after building a short–term prediction model using a single neural network. For the prediction process, statistics are collected on signal power levels and simultaneously connected users to one access point, statistical parameters of traffic. Signal power levels can also be measured using built–in wireless communication devices.

Based on the collected data, the process of training neural networks takes place. The neural network predicts the number of connected devices to one access point and possible traffic. The proposed algorithm is shown in Figure   3:

The most suitable neural network for predicting such results is the recurrent neural network. This is the most complex kind of neural networks in which feedback is available [2] (feedback refers to communication from a logically more distant element to a less distant element). The presence of feedbacks allows to remember and reproduce whole sequences of reactions to one stimulus. In terms of programming, such networks have the equivalent of cyclic execution, and in terms of systems, such a network is equivalent to a finite machine.

Consider the example at the stadium Rostov Arena [13]. According to FIFA, Wi–Fi, the network in the stands must provide access to at least 15 % of the total number of all viewers in both ranges (2,4 and 5 GHz) and at Internet access speeds of at least 1 MB/s. Selected access point model & ndash; Cisco AIR & ndash; LAP1142N–A–K9. The selected modedl provides the use of 802.11n technology in commonly used industry and is powered by 802.3af Power over Ethernet.

Based on the maximum number of possible visitors at the stadium, which has a capacity of 45000  Determine that the required number of APs for securing the Interenet network will be 350 devices. Next, knowing the power consumption & nbsp; & ndash; About 15,4  W, you can calculate the economic costs of energy supply. To do this, use the formula:

Where W – power consumption (kW);
T – operating time (2 h);
S – Tariff equal to 1 Wh = 6 rubles.
By calculations  [12],The necessary number of ATs to ensure access to the Interenet network during the match with full employment of seats will be 350 pcs. This number during 2–hours of work will consume 4,804 kW. Next, we will take the statistics of attendance of this stadium during 2019 & nbsp; years. There were 18 home matches held during this period, taking place at the Rostov Arena Stadium. Data on the number of seats occupied at these matches and the approximate requirement of access points to service 15 & nbsp;% of visitors and the cost of supplying this equipment are shown in Table 1.

Having analyzed the table, you can see that it is not always economically feasible to keep all available access points on during the match. Further, the work will consider an attempt to train a recurrent neural network to analyze the number of tickets purchased for a particular match and predict the required number of required included network equipment to access the network and independently manage them.

5. Conclusion

The above algorithm allows you to build a possible forecast model of network congestion to change the number of enabled access points, which makes it possible to increase the efficiency of controlling wireless network devices, respond in advance to changes in network structure, and provide bandwidth for traffic with the least loss and delay to critical types of traffic.

References

1. Беспроводные технологии [Электронный ресурс]. – Режим доступа: wikipedia.org
2. Нейронные сети [Электронный ресурс]. – Режим доступа: statsoft.ru...
3. Комашинский В. И., Смирнов Д. А. Нейронные сети и их применение в системах управления и связи. – М.: Горячая линия – Телеком, 2003.
4. Галушкин А. И. Нейрокомпьютеры в разработке военной техннники США. – Зарубежная радиоэлектроника, 1995 №6, с. 4–21.
5. Галушкин А. И. Современные направления развития ннейрокомпьютеров в России. – Зарубежная радиоэлектроника. Успехи современной радиэлектроники, 1998, №4, с. 3–17.
6. Atsushi H. ATM communications network control by neural network. – IJCNN. Washington DC, 1089, vol. 1, pp. 259–266.
7. Brown T. X., Neural network for switching. – IEEE Comm. Mag., Nov., 1989, pp. 72–81.
8. Brown T. X., Lui Kuo–Hui, Neural network design of a Banyan network controller. – IEEE J.–Selec, Areas Commun., 1990, vol. 8, pp. 1428–1438.
9. Trouded Terry. P., Walters Stephen M., Neural network architecture for crossbar switch control. – IEEE Transactions Circuits and Syst., 1992, vol. 28, pp. 1177–1178.
10. Funabiki Nobuo, Takefuuji Yoshiyasu, Lee Cuo Chun, A neural network model for traffic control in multistage interconnection networks. – IJCNN–91, Seatle, Wash., July 8–12, 1991, vol. 2, p. 898.
11. Технологии современных беспроводных сетей Wi–Fi: учебное пособие / [Е. В. Смирнова, А. В. Пролетарский и др.] ; под общ. ред. А. В. Пролетарского. – Москва, Издательство: МГТУ им. Н. Э. Баумана, 2017. – 446.
12. Мохамед Саид Газал, В. С. Котик, А. В. Горбенко, О. М. Тарасюк. Метод проектирование беспроводных компьютерных сетей с учетом требований к пропускной способности. Национальный аэрокосмический университет им. Н. Е. Жуковского ХАИ, Харьков 2009 г.
13. Ростов Арена – Википедия [Электронный ресурс]. – Режим доступа: wikipedia.org