Research On Key Techniques Of A Software Radio Communication Reconnaissance Receiver

In recent years, with the development of war theories and the varieties of war status, the environment of communication reconnaissance is becoming more and more complicated. High developed communication techniques of advanced electronic jamming and anti-jamming demand much higher performance of a communication reconnaissance system. In order to meet the challenges above, improving the performance of a communication reconnaissance system has always received much attention in electronic countermeasures (ECM), also known as electronic attack, which is a component of electronic warfare (EW). Communication reconnaissance receivers are main parts of a communication reconnaissance system, which can search, intercept, measure, analyze, recognize and surveil wareless communication signals in order to obtain the technical parameters of receiving signals, positions of radiant points and other relative information. With the high development of communication reconnaissance techniques, it is a trend that software defined radio (SDR) is applied in communication reconnaissance receivers. In this dissertation, some key techniques of a software radio communication reconnaissance receiver are studied to improve the performance of the receiver and satisfy the requirements of interception and reconnaissance in ECM.Firstly, after analyzing smart antennas based on the SDR techniques, a model of a communication reconnaissance receiver based on the smart antenna structure is proposed, which can process signal modulated in time domain, frequency domain and space domain with good performance of multibands, multimodes and easy updating. After analyzing each modules of the receiver, the channelized processing module and the polyphase filter algorithms are studied and simulation and analysis for real-signals are carried out in the channelized module. In addition, the design and applications of smart antennas in the communication reconnaissance receiver are researched. Then, the CAB blind beamforming algorithms based on cyclostationary signals are also studied and simulated. Simulation results show that with the directional performance of smart antennas, the proposed receiver not only can intercept and capture all signals in a wide frequency band but also can implement direction finding.Secondly, since spread spectrum communication techniques are widely used in communication reconnaissance field, it is important for communication reconnaissance receivers to detecte spread spectrum signals and estimate their parameters. Hence, in this dissertation, new blind estimation methods of PN code period in a direct sequence (DS) system are proposed. For short code signals, a modified cepstrum method based on the conventional spectrum estimation algorithms is proposed, which improves the traditional cepstrum method by using a bias autocorrelation function to estimate the power spectrum and then windowing this function and revising the spectrum. Simulation results show that the modified cepstrum method is better, its peak value is more extrusive and more smooth which is beneficial to peak searching, and the limit of signal to noise rate (SNR) achieves 2.1dB better than the traditional cepstrum method. Another modified cepstrum method for short code signals is based on the modern spectrum estimation algorithms, in which Yule-Walker or Burg algorithms are used to estimate the power spectrum, and then the cepstrum is adjusted. In addition, rules to choose the order of AR model in cepstrum method is concluded via plenty of simulation. Simulation results show that the modified cepstrum method based on the modern spectrum estimation is better than the one based on the conventional spectrum estimation. For long code signals, a modified cepstrum method based on delay and multiply preprocessing is proposed after analyzing m sequences, in which the receiving signals are delayed and multiplied by the original signal, and then the modified cepstrum method for short code signals is used to estimate the PN code periods of long code signals. Simulation results show that the modified method is an efficient approach to estimate the PN period in a long code communication system.Then, in order to recognize the modulation types of the receiving signals, including narrow-frequency-band signals and despread signals, features are selected to form a feature vector for the following classifier after feature extraction. Four methods of feature extraction are used to form the original feature set, including those methods based on instantaneous information, wavelet analysis, high order cumulants and fractal theories. Then, the genetic algorithm (GA) is applied to select an appropriate feature vector as the input of the following classifier after researching the theory, basic manipulations, operation procedures and characteristics of GA. Then, the selector based on GA for feature selection is designed, and the most optimal features are selected from the original feature set to form the feature vector, which can depress the dimension of feature space. Computer simulation by a discrete wavelet neural network (DWNN) classifier is carried out and the simulation results show that GA can improve the performance of signal modulation recognition by reducing the dimension of the feature space and selecting different feature vectors according to the different signal groups.Lastly, after researching and analyzing DWNN and adaptive resonance theory (ART) neural networks, a classifier based on the combined neural netwoks of ART2A-DWNN is proposed to classify or recognize the modulation types of the receiving signals. DWNN classifier can converge fast without local minimum values, and be robust to noise with high recognition accuracy rates. But, the DWNN classifier neither can be extended nor recognize too many modulation types at one time. However, ART neural networks can easy be extended. The classifier proposed in this dissertation combines DWNN and ART2A-E neural networks, in which ART2A-E neural networks are the first layer of the classifier to classify signals into groups coarsely, and DWNN are the second layer to classify signals in each group concretely. Each group after coarse classification has less signal types in it, which can make DWNN classifiers work efficiently. Simulation results show that the ART2A-DWNN classifier can recognize more modulation types with high accuracy rates, be extended, and be very robust to noise with a high processing speed. It is proved that ART2A-DWNN classifiers are more advanced than DWNN classifiers or ART classifiers.