Signal Modulation Recognition Based On ANN Optimized By CCGA

Signal modulation pattern recognition is very important of non-cooperative communication and it is the essential technology of the administrative department of civil radio spectrum management and military electronic countermeasures. The concept of software radio and cognitive radio was put forward, and with the development of microelectronics technology, which makes the multi-institutional communication system in the same receiver as possible, that put new requirements on the modulation of the signal pattern recognition. This thesis is on the basis of previous research to use the neural network optimized by CCGA(cooperative co-evolutionary genetic algor-ithms)to comply the signal modulation pattern recognition based on statistical model.Firstly, the two ways of the signal modulation pattern recognition based on decision theory and statistical modeling are described in the thesis. The parameters extracted of instantaneous are the basis of statistical pattern recognition, but the instantaneous parameters extracted is lack and with restrictions by analytic signal method, so the HHT(Hilbert-Haung Transform) method is used to extract the instantaneous parameters. The results of simulation show that the HHT method can overcome some of noises.Secondly, the parameters extracted as the feature vector to recognize the single signal modulation pattern using ANN(Artificial Neural Network) is described in the thesis. The basic principles and pattern recognition methods of ANN is analyzed briefly, then the BP neural network is applied to the signal modulation pattern recognition. Against slower convergence and easy to fall into local minimum point of BP neural network, the genetic algorithm (GA) is used to optimize the neural network structure and connection weights. The results of experimental show that the neural network optimized by GA can effectively improve the network performance, the classification time shortened by about60%than BP network and nearly40%shorter than the LM network with the recognition rate raised. GA emphasis on competition of survival too much, and ignore the cooperative relationship between the individual. The neural network optimized by CCGA is focused. The sub-population division, sub-individual choice, sub-individual encoding, sub-individual crossover and mutation are discussed detail in the thesis. The results simulated show that, with the same conditions, the neural network optimized by CCGA is significantly better than neural network optimized by GA, and its running time shortened by40%with recognition rate raised.Finally, the ANN optimized by CCGA is used for multi-signal modulation pattern recognition. Separation of multi-signal on the receiving model is to determine the signal separated modulation pattern. With the noise and filter performance, the time-domain characteristic parameters of the signal separated are used to determine the modulation model, the recognition rate is lower. In order to improve recognition performance, the short-term average center frequency and short-term average bandwidth are extracted by AR(AutoRegressive) model and the dimension is compressed by histogram as the feature vector to fix the modulation pattern, the recognition rate raised about12%.