Simulation and modeling of sensor networks with competitive neural networks

Academic Thesis to obtain a Masters Degree: Computer Software Trend

Abstract

In a sensor network that is a distributed distributed system, one of the issues discussed is communications synchronization. One of the main tasks of synchronizing processes is mutual exclusivity. The new algorithms provided are more fair compared to the old algorithms. In this thesis, we present a model using competitive neural networks for distributed mutual exclusivity. It is shown that the time labels, execution time and other effective parameters are predicted by competitive neural networks and the model can analytically solve the problems that occur in the critical region. The model can be simulated using Hamming and Hopplefield methods to predict its effects and its speed and accuracy graphs can be analyzed. The described model can reduce system information and is compatible with priority learning systems. Therefore, it is possible to use competitive neural networks as a model of distributed system to optimize reliability, fault tolerance and access to mutual exclusivity and critical area management. Therefore, the presented new method increases fault tolerance and centralized and distributed algorithms can use it, and thus reliability increases.

Key words: sensor networks, distributed system, critical region, mutual dependence, competitive neural network, simulation, modeling

Chapter 1- Introduction

Any system that is based on a set of machines that has memory They are not shared, implemented and run for users as if they were on a computer, a distributed system. A group of these systems are pervasive distributed systems that, unlike other types, do not have fixed nodes and permanent and high-quality communications. An example of pervasive distributed systems is sensor networks, which often use a large number of nodes for monitoring and measurement applications. In distributed systems, one of the topics discussed is simultaneity and synchronization, and usually the discussions are about the logical synchronization of the processes and the order of the correct execution of the processes. One of the most important processes in the discussion of synchronization is the access of different processes to the same variables and shared memory (critical area), which is known as mutual exclusivity. One of the challenges of the algorithms presented in this field is the discussion of establishing justice between processes and not dealing with deadlock and starvation.  The challenges of the presented algorithms in this field either work on a specific distribution or do not have a suitable diversity in dealing with different problems.

Since the neural networks themselves are a non-linear and distributed model. In this thesis, we will solve the problem of access to the critical region and mutual monopoly by using competitive neural networks, and by modeling each process with a neuron cell and each resource in the critical region with a resource in neural networks, we try to solve the mutual monopoly in distributed systems. Due to the need to have only one process in the critical region and the use of competitive neural networks, a model of these networks that has only one winner (maximal network, Mexican hat, and Hamming) will be considered against clusters (self-organizing Cohen maps and vector quantization learning). We will also discuss issues related to fault tolerance, reliability, justice in access to the critical area according to the presented model.

In order to check the ability and acceptability of the presented method, we will try to use existing data codes. The number of accesses to the critical area and the change in this number of accesses will try to prove the scalability and reliability of the method.

The linear and non-linear functions produced will be used as input to analyze the presented method. The criterion for showing the efficiency and acceptability of the presented method is the comparison of execution speed.Of course, in the comparison of execution speed due to the lack of access to a suitable criterion for measuring CPU times, the number of functions called against the number of requests will be used to reach the critical area in each of the executions, and the criterion will be defined based on it. Research that lasted from the mid-1970s to the early 1990s created a conceptual framework and algorithmic basis that proved to tolerate any value in any task consisting of two or more computers connected in a network (mobile or fixed, wired or wireless, distributed or pervasive). This type of awareness covers many areas that are the foundation of pervasive computing. Along with these types of systems, the following concepts have also been proposed [3]:

Remote communication: including protocol layering, calling remote procedures, using timeout[1] and using end-to-end arguments in placing operations[2].

Fault tolerance: including indecomposable transactions[3], distributed and nested transactions[4] and the two-phase commit protocol (A protocol for coordinating changes to recoverable resources when more than one resource manager is used by a transaction.)

High availability: includes optimistic and pessimistic replication control, mirroring, and optimistic recycling.

Remote data access: includes caching[5], function portability[6], distributed file systems, and distributed databases.

Security: includes authentication. Two-way identity and privacy based on cryptography. 1-2- Sensor networks Data transmission networks [7] can be divided into two major categories, wired[9] and wireless[10], based on their physical layer[8]. In wired networks, network nodes are connected through a visible physical device such as Ethernet or fiber optic cable. However, in wireless networks, there is no such connection device and the communication of nodes is established through radio communication or infrared rays. This classification is important because due to the specific type of physical layer used in wireless networks, they can be compared with wired networks in terms of parameters such as interference [11], reliability [12], available energy sources and security level [4].

Another division for networks is based on the ability of nodes to move in the network. In some networks, the location of the nodes is fixed, so the topology and structure of the network does not change. Unlike fixed networks [13], in mobile networks [14] the location of nodes and thus the structure of the network can be changed. Fixed or mobile network has nothing to do with the type of physical layer. For example, a mobile phone is an example of a mobile wireless network; On the other hand, a building that uses a wireless network is an example of a wireless network, but it is fixed [5]. Wireless networks (fixed or mobile) can be considered in two ways: wireless networks that have infrastructure [15] and wireless networks without infrastructure. In the first type, network nodes are connected (wirelessly) to higher-level nodes such as base stations [16] or access points [17] and thus will be able to communicate; While in the second type, there are basically no such nodes and the network is able to start its work without any basic infrastructure, and the nodes are self-initiated[18] [6] [.

Ad hoc networks[19] are a special type of wireless networks in which nodes try to communicate with each other without any fixed infrastructure. The flexibility of these networks allows us to launch them anytime and anywhere. Among the most important applications of contingency networks, we can mention the communication between forces and equipment in a battlefield (as the beginning of contingency networks also goes back to the project of DARPA [20] and the US Department of Defense), relief operations in unexpected incidents and even multi-user games. Another type of wireless networks are sensor networks [21]. A sensor network is a collection of many nodes that work together to collect information and transmit it. Rapid detection of forest fires and detection of moving objects in a region (for example, a border) are examples of sensor networks [7].

In many cases, it is necessary to provide a level of security in contingency and sensor networks.