Aplicação de Redes Neuronais de Competição à Desmodulação de Sinal em Modulação Digital

This work applies a pattern recognition system based on artificial neural networks to the demodulation of digitally modulated signals. It is intended to explore the ability that may be displayed by neural networks to reconstruct patterns corrupted by noise. A competition neural network is presented that is able to perform digital modulation reception tasks: frequency and phase carrier synchronization; symbol synchronization; systematic compensation of symbol rotation. Some tasks imply post-processing of neural network output. The presented system does not require carrier synchronization with a local oscillator. Same details of the proposed methods are new contributions, which are stressed upon discussion. It was indented to guarantee future implementation of the neural network using programmable logic arrays. It is necessary to limit the proposed architecture size to an acceptable number of logical gates. On the other hand it is necessary to guarantee that each given stimulus vector is fully processed before the next one arrives (assuring real time operation). This work is mainly focused on 4-PSK (or QPSK) modulation, since this kind of modulation is used in a wide range of digital modulation applications (including third generation mobile phones). Generalization possibilities to M-PSK and M-QAM are discussed. Simulation methods were applied to evaluate system performance. Results indicate that the system is able to perform accordingly to objective