Research For Modulation Recognition Based On Improved Ant Colony Algorithms And Neural Networks

modulation recognition technology in the field of communications has been knowledge as an important position in the wireless application .More and more scholars focus on this research. Militarily, mainly for surveillance, reconnaissance and communications, electronic warfare and threat analysis, In the civilian, for spectrum management, spectrum monitoring, interference identification, such as radio signal positioning management.Classification is the key characteristics of Modulation technique, The current technology used for the classification of the main characteristics is neural network classifier . Neural Network Classifier can be good pattern recognition tool because of its strong ability to high recognition rate , automatically adapting to environmental change, better to dealing with complex nonlinear problems, having better stability and potential fault tolerance. Traditional neural network classifiers commonly be trainedused with BP algorithm, and the BP algorithm as a gradient algorithm, it can not guarantee the right of the value of connecting to the global optimal solution convergence. And the convergence is too slow and local, which in effect weakened the Neural Network Classifier advantage.Therefore, the main research in this article to improve classification performance and training speed of neural network classifiers, and the robustness of the key modulation recognition.First of all, the paper proposes a algorithm based on the elite and optimization to sort in the ant colony. Apart from the basic ant colony algorithm randomized, parallel search features, all search path will be sorted, in accordance with the results of each search path f such as the length of its path. The extra pheromone will be added to the excellent path according to the sequence of sort . In this way, the more higher ranking the path of the ants in the search, more information will be added, and the more ants gathered to them.Compared to the basic ant colony algorithm ,the improve ant colony algorithm markedly be improved in convergence rate, but still flawed, that is in the initial stages of ant algorithm, due to lack of early plaque pheromone, the ant colony algorithm for the time complexity expand for fast growth in time, slow to solve.In order to further improve the algorithm convergence rate, this paper presents a hybrid genetic ant colony algorithm, combining genetic algorithms and improved ant colony algorithm. The first step , generate the initial pheromone distribution by the characteristics of the use of genetic algorithms, rapid and comprehensive manner, then find the optimal solution to a problem with the positive feedback of ant colony algorithm. The algorithm convergence speed is superior to genetic algorithm and ant colony algorithm, with a higher accuracy for searching optimal solution.Finally, the hybrid genetic ant algorithm will be used for training the weights of neural networks as classifier of charistrics of modulation signal. The method is according to the purpose of classification , using the advantages of non-linearity and adaptiveness of neural networks , and combining with the algorithms of fast training, robust and global convergence. It overcomes the drawbacks of the general classifier of neural networks. Computer simulations indicate good performance on an AWGN channel, even at signal-to-noise ratios as low as 5 dB. This compares favorably with the performance obtained with most algorithms based on pattern recognition techniques.