A cognitive signal recognition and demodulation model is designed based on the cyclic neural network for conventional amplitude-phase satellite modulations. Through the cyclic neural unit

the features of the target signal are extracted. And the features are dimension-mapped by the fully connected neural network. The model finally completes the modulation recognition and demodulation of the target signal with these mapped features. This method does not need much prior knowledge about signal-to-noise ratio (S/N)

and it is not sensitive to frequency offset. The method also has good adaptability in the maintenance and extension

which conforms to the demands of the engineering

while the traditional algorithms need to redeploy the decision rule. Computer simulations show that the correct recognition probability is close to 98% when S/N is greater than 6 dB and demodulation error rate is close to the theoretical gate. The presented theoretical form provides a new idea for intelligent signal processing

and it can also be used in other communication signal processing fields.