Automatic Recognition Of Satellite Signals Modulation Based On Wavelet Neural Network

The transmitting message all is carries on by the different modulation type in the different frequency channel. Along with the diversification and the complication of the communication system and the modulation style, the signal environment is more and more crowded. In many applications, usually needs to distinguish and to monitor these signals, for example civil signal authentication, disturbance recognition and frequency spectrum management. The civil authority possibly must monitor the civil signaling, in order to retain one's hold or to discovery unregistered transmitter. Other applications are military goals, for example radio intercept, the digital signal reconnaissance, the electronic countermeasure, and the threat analysis and so on all involve to the signal modulation recognition.Along with the daily increase of people's demand for the real-time service, satellite communication fast develops into the most prosperous aspect of the communication domain by its large coverage area, far transmitting range , wide communication band, stable transmit connection and so on the merit. Before we demodulate the receiving signal from the satellite, we must determine the modulation way of the signal and its correlation parameter like carrier frequency, signal band width, symbol speed rate and so on. Former observation and control system is in view of sole modulation way. The ground control station is waiting to receive the signal with the known modulation way while the satellite is passing by. There is no need to recognize the modulation way.After has introduced the auto-adapted channel real-time processing technology, the signals are able to choose suitable modulation way along with the channel condition, which request the receiving apparatus can automatic diagnosis the correct demodulation way in order to adopt. Moreover, to enable a control station to carry on the observation and control of many satellites, the traditional method needs to dispose the multi-wraps equipment and use the manual change. The cost is high; the operation is complex, uses the automatic modulation recognition can solve these problems. Similarly, in order to construct the global observation and control network, also needs the equipment on the satellite to have the automatic signal modulation recognition ability. Because one satellite will pass through different ground stations when it circum around the earth. Obviously, to study the modulation recognition algorithms and find highly effective and accurate modulation recognition algorithms is extremely important.In recent years, many universities and institutes have pay great attention to the automatic modulation recognition algorithm. The research has got some development, and the embryonic system has been formed. At present, the recognition methods approximately divide into two big kinds: the likelihood method based on the decision tree and the pattern recognition method which withdraws based on the characteristic. Former method is a test question based on many hypotheses, its characteristic is to supposes some candidate modulation way by observing the identification signal then judges the similarity and determines its modulation way; Latter method must withdraw the characteristic vector through the characteristic extraction system from the receive signal first, then determines the modulation type of the signal through the pattern recognition system. The decision-making theory method is in the decision-making theory frame, studies the question using the probability and the supposition examination viewpoint. Through analysis of the signal likelihood function, obtains sufficient statistic which uses to classify, then choose an appropriate threshold for comparison to complete the automatic recognition. Uses this method in the smallest average cost function significance is most superior. But this method need more prior knowledge, like distribution function and average value, variance as well as signal to noise ratio parameter and so on. Moreover, its formula is complex and the computation quantity is big, so this method is difficult for real-time processing. Therefore, today the pattern recognition method which withdraws characteristic is popular, and researches are all concentrate in this method.This article analyzes the existing modulation recognition algorithm, consider the characteristics of satellite transmit channel, and propose the wavelet neural network to complete the satellite modulation classification. This article has analyzed modulating characteristic of each modulation way, used the heterogeneity filter structure to get the characteristic, has given elaboration of the instant characteristic of the signal in time domain, has exhaustively analyzed two kind of classical characteristics vectors - statistical characteristic quantity and the high order cumulant. Then, the article introduced the most formidable processing tool– neural network, simultaneously introduced the power value wavelet nerve network for classification, and has given the detailed elaboration to the wavelet neural network in terms of its architecture and training algorithm. Finally, according to the satellite signal, withdraws the suitable characteristic parameter, carries on the classified recognition using the wavelet neural network, the simulation experiment obtained the better recognition effect. Main work as follows:1. Principle and algorithm of the statistical pattern modulation recognition. The statistical pattern recognition method is comes by the classical pattern recognition theory which include the characteristic withdraw part and the classified part. The characteristic withdraw part gets useful information from the primary data. It is one kind of mapping relations, namely maps from the input signal sequence to the designated characteristic space. The main function is to judge the signal modulation type the subordinate relations. But the statistical pattern recognition method usually are based on the non- noise jamming supposition. If for great signal to noise ratio situation, characteristic obviously, is easy to withdraw, can have the better recognition performance; But when low signal to noise ratio, characteristic is fuzzy, then difficulty with withdraws, the recognition effect is worse.2. Summarizes the characteristic parameter proposed by the predecessor The characteristic withdraws is essential part of the pattern recognition. It extracts useful information from the input signal sequence. The main goal is to attribute the remarkable category difference as far as possible. Another goal is to reduce the data set and enhance the recognition efficiency. This article proposed six especial parameter, they are: Normalized instantaneous amplitude maximum valueγmax; Standard deviation of zero center normalized instantaneous amplitude absolute valueσaa; Standard deviation of zero center instantaneous phase absolute value for non-linear component of un-weak signalσap; Standard deviation of zero center instantaneous phase for non-linear component of un-weak signalσdp; Standard deviation of zero center instantaneous frequency for un-weak signalsσaf and Ratio P; This method is and does not need any prior knowledge.3. Wavelet neural network modeling. With the development of the artificial neural network research, its application expands day by day. The neural network has the information distributional memory, the auto-adapted parallel processing and high fault-tolerant ability. These characteristic is the foundation for pattern recognition. Neural network may use in the pattern recognition is include: BP network, RBF network, ART network, high order network, Hopfield network, wavelet neural network and so on. BP network has been used popular since long ago, because its theory development is mature, and the network architecture is clear. But the BP network has several shortcomings one is the partial minimum problem; the other is the slow convergence rate. The wavelet network appearance has made up this kind of insufficiency in the certain degree.4. Design the power value wavelet neural network for modulation recognition. In the experiment, we select the wavelet neural network which has only one concealed level, suppose the nodes of the concealed level are10. Here take above 6 characteristic parameters as the input vector. The network may distinguish six modulations ways, therefore the output vector is 6, and the network uses the BP algorithm to go on the training.5. Simulation and analysis of the recognition system.We make simulation under the circumstance of signal to noise ratio is 5db,10db,and 20db ,the overall success ratio is not smaller than 95%,and the stability of the recognition is good.The algorithm proposed by this article is simple and effective. It can be used under the lower signal to noise ratio environment to recognize the satellite communication signals. It provided basis for the automatic modulation recognition; also laid good foundation for further blind examination technology.