A.V. Eltishev - Student, Department of Automation and Telemechanics, Perm National Research Polytechnic University
А.I. Posyagin – Assistant, Department of Automation and Telemechanics, Perm National Research Polytechnic University. E-mail: email@example.com
А.А. Yuzhakov – Dr.Sc. (Eng.), Professor, Department of Automation and Telemechanics,
Perm National Research Polytechnic University
The given article concerns the neural network of the self-routing analog-to-digital converter in which emphasis is put on a simplified structure consisting of basic neurons connected in circle. This simplification made it possible to develop an imitating model of a neural network using the routing algorithms with no consideration of the devices providing them. Therefore, this model has the following limitations: the probability of other elements\' failure, apart from the basic neurons, is not considered; the model has the simple-connected structure, so the failure of a single neuron leads to the loss of the circle structure and the decrease of the functional capacity to allocate individual analog-to-digital converters in the neural network. This imitating model was examined for adequacy with the help of the created analytic model in the form of the queuing network. Since the model proved to be adequate it helped to pursue the research of the probability of failure to serve the request of forming the analog-to-digital converter which comes to the neural network. The probability of such failure is one of the key parameters for the neural network analog-to-digital converter because it constitutes the real-time system, therefore, no signal delays have to occur. At the current stage of modeling there were discovered the relations between the probability of failure and other parameters of the neural network such as: the amount of inputs, the amount of the basic neurons given, minimal and maximal capacity of the individual analog-to-digital converters formed, the intensity of the input stream of requests and also the intensity of basic neurons\' failure stream in the network. These relations allowed to single out the most important parameters of the network which influence the security of the neural network analog-to-digital converter\'s failure-resistance. In what follows, it is planned to continue the investigation of this model combining various parameters in order to reveal common patterns in changing the probability of the failure to serve the request. Moreover, it is planned to update this model in order to receive the opportunity to add a random amount of connections and also to consider the probability of failure of other network elements apart from the neurons.