| Autor: |
Nikolay Neudobnov, Evgeny Efimov |
| Rok vydání: |
2021 |
| Předmět: |
|
| Zdroj: |
2021 Systems of Signals Generating and Processing in the Field of on Board Communications. |
| Popis: |
This paper reviews a subset of approaches to the angle-of-arrival problem that can be redefined in a form compatible with the techniques based on artificial neural networks. That subset includes a step viewed as a mapping between a set of cross-delays and the estimated value of the angle of arrival. The paper demonstrates artificial neural networks to be a suitable solution to the particular part of a problem. The conventional multilayer perceptron and specialized angular network topologies are presented to emphasize the essential step of first considering a prior knowledge of the problem while synthesizing and training a network. The numerical simulation results including the plot of the loss function and the trainable parameters gradients during the training process are presented for both models. The results indicate that the angular model appears to be superior to the conventional one. Additional note and simulation for the angular model is performed to reveal the generalization ability of a properly trained neural network. This makes it possible for the network to withstand a certain level of noise that contaminates the input data. Some important points, which left outside of the scope of the main part of the article, and key points for further research are highlighted in the conclusion. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
